Fetal brain MRI tissue analysis method, device and electronic equipment

Abstract

The present application relates to the field of data analysis, and in particular to a fetal brain MRI tissue analysis method, apparatus, and electronic device, comprising obtaining multiple post-treatment brain MRI images, each corresponding to a different gestational age; performing brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues; obtaining, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map; determining, for each post-treatment brain MRI image, a deformation coefficient for each local brain tissue; generating, for each local brain tissue, a linear analysis map corresponding to each local brain tissue based on the deformation coefficients of the local brain tissue corresponding to all gestational ages; and determining final brain tissue analysis information based on the linear analysis maps corresponding to all local brain tissues. The present application facilitates a holistic analysis of the impact of changes before and after treatment on different gestational ages and different brain tissues.

Description

Technical Field

[0001] The present application relates to the field of data analysis, and in particular to a fetal brain MRI tissue analysis method, device, and electronic device. Background Art

[0002] Congenital diseases are diseases acquired during the fetal period, that is, the fetus is affected by external or internal adverse factors during its growth and development in the womb, resulting in abnormal development. Generally, there are manifestations or signs at birth. Once a congenital disease occurs, it will bring a serious burden to the family and society. Whether the child dies or survives, it will cause great pain to the child and the family. Timely detection and treatment of congenital diseases in the fetus can help reduce the mortality rate. Therefore, regular prenatal maternal and child health examinations are particularly critical to conduct regular health analysis of the developing fetus.

[0003] Spina bifida is one of the most common congenital diseases and can impair fetal brain development. Corresponding brain tissue changes include hypoplasia of the cerebellar vermis and elongation of the fourth ventricle. Related technologies facilitate fetal health examinations and analysis by viewing fetal brain tissue information through ultrasound images, allowing diagnosis and treatment of congenital diseases based on brain tissue changes. However, this technology fails to comprehensively analyze the impact of different gestational ages and brain tissue changes during treatment, hindering analysis of postnatal outcomes. Summary of the Invention

[0004] In order to facilitate an overall analysis of the effects of changes before and after treatment on different fetal ages and different brain tissues, the present application provides, in particular, a fetal brain MRI tissue analysis method, device, and electronic equipment.

[0005] In a first aspect, the present application provides a fetal brain MRI tissue analysis method, which adopts the following technical solutions:

[0006] A fetal brain MRI tissue analysis method comprising:

[0007] Acquiring a plurality of post-treatment brain MRI images, wherein each post-treatment brain MRI image corresponds to a different gestational age;

[0008] Perform brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues;

[0009] For each post-treatment brain MRI image, obtaining a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map;

[0010] For each post-treatment brain MRI image, determining the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map;

[0011] For each local brain tissue, a linear analysis graph corresponding to each local brain tissue is generated according to the local brain tissue deformation coefficient corresponding to all fetal ages;

[0012] Final brain tissue analysis information is determined based on the linear analysis graphs corresponding to all local brain tissues, and the brain tissue analysis information is used to analyze the correlation between treatment effect and gestational age.

[0013] By adopting the above technical solution, multiple post-treatment brain MRI images of different gestational ages are obtained, and brain tissue is segmented for each post-treatment brain MRI image to obtain multiple local brain tissues, and multiple post-treatment local brain tissue segmentation maps corresponding to the multiple local brain tissues are obtained. For each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue is obtained. Secondly, for each post-treatment brain MRI image, the deformation coefficient of each local brain tissue is determined based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, a linear analysis map corresponding to each local brain tissue is generated according to the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, final brain tissue analysis information is determined based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct an overall analysis of the impact of changes before and after treatment of different gestational ages and different brain tissues.

[0014] In one possible implementation, the brain tissue segmentation of each post-treatment brain MRI image is performed using a trained pyramid segmentation model. The training process of the pyramid segmentation model includes:

[0015] Acquire multiple brain MRI training samples, each brain MRI training sample including a brain MRI image and corresponding multiple local brain tissue labels;

[0016] Obtaining a pyramid segmentation model to be trained, wherein the pyramid segmentation model to be trained includes an encoder, a decoder, a skip connection module, and a feature combination module;

[0017] Inputting multiple brain MRI training samples into the pyramid segmentation model to be trained for iterative training to obtain a training result for each brain MRI training sample;

[0018] Determine the function loss value based on each training result and the corresponding local brain tissue label;

[0019] If the function loss value is lower than the preset standard loss value, the iterative training is stopped to obtain a trained pyramid segmentation model.

[0020] By adopting the above technical solution, multiple brain MRI training samples are obtained, wherein each brain MRI training sample includes a brain MRI image and corresponding multiple local brain tissue labels, and then the multiple brain MRI training samples are input into the obtained pyramid segmentation model to be trained for iterative training, and the training results corresponding to each brain MRI training sample are output, wherein the pyramid segmentation model to be trained includes an encoder, a decoder, a skip connection module and a feature combination module, and then a function loss value is determined by each training result and the corresponding local brain tissue label. When the function loss value is lower than a preset standard loss value, the iterative training of the pyramid segmentation model to be trained is stopped to obtain a trained pyramid segmentation model. Through iterative training, the accuracy of training the pyramid segmentation model is helped to be improved.

[0021] In one possible implementation, performing brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues includes:

[0022] Obtaining post-treatment brain MRI image features corresponding to each post-treatment brain MRI image;

[0023] Input each post-treatment brain MRI image feature into an encoder for feature extraction, and output the encoded MRI image feature corresponding to each encoding layer, wherein the encoder includes multiple encoding layers, each encoding layer includes a downsampling module and an attention structure module;

[0024] Based on the decoder and skip connection module, all the encoded MRI image features corresponding to the decoding layer are feature-connected, and the target image features corresponding to each decoding layer are output. The decoder includes multiple decoding layers, each decoding layer includes an upsampling module and an attention structure module, and the last decoding layer also includes a hybrid dilated convolution module;

[0025] Multiple target image features are input into the feature combination module, and multiple post-treatment local brain tissue segmentation maps are output.

[0026] By adopting the above technical solution, the MRI image features corresponding to each post-treatment brain MRI image are first obtained, and then each post-treatment brain MRI image feature is input into the encoder for feature extraction, and the encoded MRI image features corresponding to each encoding layer are output, wherein the encoder includes multiple encoding layers, and each encoding layer includes a downsampling module and an attention structure module. Then, based on the decoder and the jump connection module, all the encoded MRI images corresponding to the decoding layer are feature-connected to obtain multiple target image features. Finally, the multiple target image features are input into the feature combination module for combination to obtain multiple post-treatment local brain tissue segmentation maps, thereby enhancing the accuracy of local brain tissue segmentation and thereby improving the segmentation accuracy.

[0027] In one possible implementation, the step of inputting each post-treatment brain MRI image feature into an encoder for feature extraction and outputting the encoded MRI image feature corresponding to each encoding layer includes:

[0028] Input each post-treatment brain MRI image feature into a first encoding layer, and output a first encoded MRI image feature;

[0029] The first encoded MRI image feature is input into the second encoding layer to obtain the second encoded MRI image feature, and the loop is executed until the last encoding layer in the encoder outputs the last layer of encoded MRI image features to obtain the encoded MRI image features corresponding to each encoding layer.

[0030] By adopting the above technical solution, each post-treatment brain MRI image feature is input into the first coding layer, the first coded MRI image feature is output, and then the first coded MRI image feature is input into the second coding layer to obtain the second coded MRI image feature. The above steps are repeated until the last coding layer in the encoder outputs the last layer of coded MRI image features, and finally the coded MRI image features corresponding to each coding layer are obtained. The attention structure model is added to the encoder to improve the accuracy of extracting the coded MRI image features.

[0031] In one possible implementation, the decoder and skip connection module performs feature connection on the encoded MRI image features corresponding to all decoding layers and outputs the target image features corresponding to each decoding layer, including:

[0032] Inputting the encoded MRI image features output by the last layer in the encoder into the first decoding layer to obtain first decoded MRI image features;

[0033] Based on the jump connection module, the first decoded MRI image features and the encoded MRI image features of the corresponding encoding layer are input into the second decoding layer to obtain the second decoded MRI image features. The loop is executed until the last decoding layer in the decoder to obtain the target image features corresponding to each decoding layer.

[0034] By adopting the above technical solution, the encoded MRI image features output by the last layer in the encoder are input into the first decoding layer to obtain the first decoded MRI image features, and then based on the jump connection module, the encoded MRI image features of the corresponding encoding layer are obtained to complete the splicing, and the splicing result is input into the second decoding layer to obtain the second decoded MRI image features and the corresponding target image features, until the last decoding layer in the decoder outputs the corresponding target image, so as to obtain the target image features corresponding to each decoding layer. The combination of the encoding process and the decoding process is realized through the jump link module, which helps to improve the accuracy of obtaining the target image features.

[0035] In one possible implementation, the attention structure module is used to:

[0036] Obtaining pixel feature values ​​of the sampled image, when the attention structure module is the attention structure module of the encoding layer, the sampled image is a downsampled brain MRI image after treatment; when the attention structure module is the attention structure module of the decoding layer, the sampled image is an upsampled encoded MRI image corresponding to the decoding layer;

[0037] Determine the corresponding matrix Q, matrix K and matrix V according to the pixel eigenvalue;

[0038] Convolving the matrix K to obtain a matrix K1;

[0039] Concatenate the matrix K1 and the matrix Q to obtain an initial matrix, and perform two consecutive convolution operations on the initial matrix to obtain a relational matrix A;

[0040] Perform matrix multiplication on the relationship matrix A and the matrix V to obtain a matrix K2;

[0041] The matrix K1 and the matrix K2 are fused to output image features, wherein when the attention structure module is the attention structure module of the encoding layer, the encoded MRI image features are output; when the attention structure module is the attention structure module of the decoding layer, the target image features are output.

[0042] By adopting the above technical solution, the attention structure model is applied in both the encoder and the decoder. When the attention structure module is the attention structure module of the encoding layer, the sampled image is the upsampled brain MRI image after treatment. When the attention structure module is the attention structure module of the decoding layer, the sampled image is the downsampled encoded MRI image corresponding to the decoding layer. After obtaining the corresponding image feature values, the corresponding matrix Q, matrix K and matrix V are obtained. By performing correlated convolution, splicing, multiplication and fusion on the matrix Q, matrix K and matrix V, the image features are finally output. When the attention structure module is the attention structure module of the decoding layer, the sampled image is the downsampled encoded MRI image corresponding to the decoding layer. By establishing contextual connections, the accuracy of the output image features is improved.

[0043] In one possible implementation, obtaining, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map includes:

[0044] The gestational age was determined based on the brain MRI images after each treatment;

[0045] According to each post-treatment brain MRI image and the corresponding gestational age, the corresponding pre-treatment brain MRI image is determined from the historical database;

[0046] Determining a plurality of corresponding pre-treatment local brain tissue segmentation maps according to each pre-treatment brain MRI image;

[0047] For each local brain tissue, a local brain tissue segmentation map before treatment corresponding to each local brain tissue segmentation map after treatment is determined.

[0048] By adopting the above technical solution, the corresponding gestational age is determined based on the post-treatment brain MRI image, and the corresponding pre-treatment brain MRI image is matched from the historical database based on the gestational age and the post-treatment brain MRI image. Then, based on the local brain tissue, the local brain tissue segmentation maps before and after treatment are matched. By traversing the historical database, the accuracy of determining the corresponding pre-treatment local brain tissue segmentation map is improved.

[0049] In one possible implementation, for each post-treatment brain MRI image, determining the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map includes:

[0050] For each post-treatment brain MRI image, each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map are imported into a pre-established coordinate system to determine the coordinates of the pre- and post-treatment boundary points of each pre- and post-treatment local brain tissue segmentation map;

[0051] For each local brain tissue, the coordinates of the change point set are determined based on the coordinates of all boundary points of each local brain tissue before and after treatment;

[0052] Determine the deformation field of each brain tissue according to the coordinates of the change point set corresponding to each local brain tissue;

[0053] The corresponding deformation coefficient of each local brain tissue is determined according to each brain tissue deformation field.

[0054] By adopting the above technical solution, each local brain tissue segmentation map before treatment and the corresponding local brain tissue segmentation map after treatment are imported into the established coordinate system, the boundary point coordinates are determined respectively, and then the coordinates of the change point set before and after treatment corresponding to the local brain tissue are determined by the boundary point coordinates, and then the corresponding deformation field is determined. Finally, the deformation coefficient corresponding to the deformation field of each local brain tissue is calculated. The deformation coefficient facilitates a comprehensive analysis of the impact of the treatment effect on each local brain tissue before and after treatment.

[0055] In a second aspect, the present application provides a fetal brain MRI tissue analysis device, which adopts the following technical solution:

[0056] A fetal brain MRI tissue analysis device, comprising:

[0057] An image acquisition module is used to acquire a plurality of brain MRI images after treatment, wherein each brain MRI image after treatment corresponds to a different gestational age;

[0058] A brain tissue segmentation module is used to perform brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues;

[0059] A corresponding image acquisition module is used to acquire, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map;

[0060] a deformation coefficient determination module, configured to determine, for each post-treatment brain MRI image, a deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map;

[0061] An analysis graph generation module is used to generate, for each local brain tissue, a linear analysis graph corresponding to the local brain tissue according to the deformation coefficient of the local brain tissue corresponding to all fetal ages;

[0062] The analysis information determination module is used to determine final brain tissue analysis information based on the linear analysis graphs corresponding to all local brain tissues, and the final brain tissue analysis information is used to analyze the correlation between the treatment effect and the gestational age.

[0063] By adopting the above technical solution, multiple post-treatment brain MRI images of different gestational ages are obtained, and brain tissue is segmented for each post-treatment brain MRI image to obtain multiple local brain tissues, and multiple post-treatment local brain tissue segmentation maps corresponding to the multiple local brain tissues are obtained. For each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue is obtained. Secondly, for each post-treatment brain MRI image, the deformation coefficient of each local brain tissue is determined based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, a linear analysis map corresponding to each local brain tissue is generated according to the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, final brain tissue analysis information is determined based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct an overall analysis of the impact of changes before and after treatment of different gestational ages and different brain tissues.

[0064] In a third aspect, the present application provides an electronic device, which adopts the following technical solution:

[0065] An electronic device, comprising:

[0066] at least one processor;

[0067] Memory;

[0068] At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute the above-mentioned method for fetal brain MRI tissue analysis.

[0069] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0070] A computer-readable storage medium includes: a computer program that can be loaded by a processor and execute the fetal brain MRI tissue analysis method.

[0071] In summary, this application includes at least one of the following beneficial technical effects:

[0072] 1. Acquire multiple post-treatment brain MRI images of different gestational ages, perform brain tissue segmentation on each post-treatment brain MRI image to obtain multiple local brain tissues, and obtain multiple post-treatment local brain tissue segmentation maps corresponding to the multiple local brain tissues. For each post-treatment brain MRI image, obtain a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue. Secondly, for each post-treatment brain MRI image, determine the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, generate a linear analysis map corresponding to each local brain tissue according to the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, determine the final brain tissue analysis information based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct an overall analysis of the impact of changes before and after treatment of different gestational ages and different brain tissues.

[0073] 2. By importing each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map into the established coordinate system, the boundary point coordinates are determined respectively. Then, the boundary point coordinates are used to determine the coordinates of the change point set of the local brain tissue before and after treatment, and then the corresponding deformation field is determined. Finally, the deformation coefficient corresponding to the deformation field of each local brain tissue is calculated. The deformation coefficient facilitates a comprehensive analysis of the impact of the treatment effect on each local brain tissue before and after treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 1 is a flow chart of a fetal brain MRI tissue analysis method in an embodiment of the present application;

[0075] Figure 2 is a schematic diagram of local brain tissue in an embodiment of the present application;

[0076] FIG3( a ) is a brain MRI image during the training process of the pyramid segmentation model in an embodiment of the present application;

[0077] FIG3( b ) is a diagram showing multiple local brain tissue labels corresponding to a brain MRI image during the training of a pyramid segmentation model in a later embodiment of the present application;

[0078] FIG3 (c) is a segmentation result diagram of the pyramid segmentation model in an embodiment of the present application;

[0079] Figure 4 It is a schematic diagram of the structure of the pyramid segmentation model implemented in this application;

[0080] Figure 5 This is a schematic diagram of the structure of the attention structure module in the pyramid segmentation model implemented in this application;

[0081] Figure 6This is a flowchart of local brain tissue analysis in an embodiment of the present application;

[0082] Figure 7 This is a schematic structural diagram of a fetal brain MRI tissue analysis device according to an embodiment of the present application;

[0083] Figure 8 It is a structural diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0084] The following is combined with Figure 1-8 This application is described in further detail.

[0085] After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0086] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0087] Congenital diseases are one of the leading causes of neonatal death worldwide. During development, the tissues of fetuses with congenital diseases differ from those of normal fetuses, especially in the brain. For example, fetuses with spina bifida have brain tissue changes such as incomplete development of the cerebellar vermis, elongation of the fourth ventricle, and cell migration and deformity. Therefore, when a congenital disease is detected in a fetus, timely analysis can help reduce neonatal mortality.

[0088] However, after a fetus is found to have a congenital disease and treated, the treatment process may affect brain tissue that was not abnormal before treatment. Therefore, in order to facilitate a holistic analysis of the effects of changes before and after treatment on different gestational ages and different brain tissues, in an embodiment of the present application, multiple post-treatment brain MRI images of different gestational ages are obtained, each post-treatment brain MRI image is subjected to brain tissue segmentation to obtain multiple local brain tissues, and multiple post-treatment local brain tissue segmentation maps corresponding to the multiple local brain tissues are obtained. For each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue is obtained. Next, for each post-treatment brain MRI image, a deformation coefficient of each local brain tissue is determined based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, a linear analysis map corresponding to each local brain tissue is generated based on the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, final brain tissue analysis information is determined based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct a holistic analysis of the effects of changes before and after treatment on different gestational ages and different brain tissues.

[0089] Specifically, embodiments of the present application provide a fetal brain MRI tissue analysis method, which is performed by an electronic device, which can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be, but is not limited to, a smartphone, tablet computer, laptop computer, or desktop computer. The terminal device and server can be directly or indirectly connected via wired or wireless communication, which is not a limitation of the embodiments of the present application.

[0090] refer to Figure 1 , Figure 1 : is a flow chart of a fetal brain MRI tissue analysis method in an embodiment of the present application, the method comprising steps S110, S120, S130, S140, S150, and S160, wherein:

[0091] Step S110: Acquire multiple post-treatment brain MRI images.

[0092] Among them, each post-treatment brain MRI image corresponds to a different gestational age.

[0093] Specifically, magnetic resonance imaging (MRI) is a new examination technology based on the principle that atomic nuclei with magnetic moments can produce energy level transitions under the influence of a magnetic field. MRI is particularly important for diagnosing medical conditions. When obtaining MRI images of pregnant women, fetal brain MRI images can be extracted and used to diagnose the health of the fetus. Post-treatment brain MRI images are obtained when a congenital disease is discovered in the fetus and the corresponding treatment is given. In order to comprehensively analyze the effects of changes before and after treatment on different gestational ages and different brain tissues, it is necessary to obtain brain MRI images of sick fetuses of different gestational ages.

[0094] Step S120: performing brain tissue segmentation on each post-treatment brain MRI image to obtain a plurality of post-treatment local brain tissue segmentation maps corresponding to the plurality of local brain tissues.

[0095] Specifically, after the fetal brain tissue is segmented, seven types of local brain tissues can be generated, wherein the local brain tissues are: cerebrospinal fluid, gray matter, white matter, ventricles, cerebellum, deep gray matter and brainstem. When performing brain tissue segmentation on the brain MRI image after treatment, a CNN convolutional neural network can be used, or a pyramid segmentation model can be used. There is no specific limitation in the embodiments of this application, as long as the brain MRI image after treatment can be segmented and the post-treatment local brain tissue segmentation map corresponding to the seven types of local brain tissues can be obtained. For ease of understanding, please refer to Figure 2 , Figure 2 is the segmentation map of seven local brain tissues after treatment, among which Figure 2 (a) is the cerebrospinal fluid segmentation map after treatment, Figure 2 (b) is the gray matter segmentation map after treatment; Figure 2 (c) is the white matter segmentation map after treatment. Figure 2 (d) is the ventricular segmentation map after treatment. Figure 2 (e) is the cerebellum segmentation map after treatment. Figure 2 (f) in the figure is the deep gray matter segmentation map after treatment; Figure 2 (g) in the figure is the brainstem segmentation map after treatment.

[0096] When using a CNN convolutional neural network for segmentation, specifically, a CNN convolutional neural network is used to perform feature recognition on the fetal brain MRI image after treatment to determine the region of interest, where the region of interest is a region containing seven types of local brain tissues, and then a CNN convolutional neural network with the same structure is used to segment the determined region of interest into seven types of local brain tissues; in the implementation method of using a pyramid segmentation model for segmentation, specifically, the pyramid segmentation model includes an encoder, a decoder, a jump connection module and a feature combination module, and by inputting the post-treatment brain MRI image into the trained pyramid segmentation model, the seven types of local brain tissues are segmented, and then the post-treatment local brain tissue segmentation maps corresponding to the seven local brain tissues are output.

[0097] After performing brain tissue segmentation on the brain MRI images after treatment at each gestational age through this step, seven post-treatment local brain tissue segmentation maps corresponding to the seven local brain tissues corresponding to each gestational age are obtained. Ultimately, seven post-treatment local brain tissue segmentation maps corresponding to the seven local brain tissues corresponding to all gestational ages can be obtained.

[0098] Step S130: for each post-treatment brain MRI image, obtaining a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map.

[0099] Specifically, the brain MRI image before treatment and the brain MRI image after treatment correspond to the same fetus of the same gestational age. In order to facilitate the viewing of whether the local brain tissues have changed before and after treatment, it is necessary to obtain the local brain tissue segmentation map before treatment that corresponds to the local brain tissue segmentation map after treatment. The local brain tissue segmentation map before treatment can be obtained by performing local tissue segmentation on the brain MRI image before treatment of the fetus, or it can be obtained by performing local brain tissue segmentation together with the brain MRI image after treatment after the brain MRI image after treatment is obtained. The time for performing local brain tissue segmentation on the brain MRI image before treatment is not specifically limited in the embodiment of the present application, as long as the segmentation can be completed before the local tissue segmentation map before treatment is obtained based on the local tissue segmentation map after treatment.

[0100] In this step, the pre-treatment local brain tissue segmentation maps of all local brain tissues corresponding to the post-treatment local brain tissue segmentation maps corresponding to all gestational ages can be obtained. After this step, the pre-treatment local brain tissue segmentation maps corresponding to the post-treatment brain MRI corresponding to all gestational ages can be obtained.

[0101] Step S140: for each post-treatment brain MRI image, based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map, determining a deformation coefficient corresponding to the deformation field of each local brain tissue.

[0102] Specifically, the deformation field is formed based on the pixel points or voxel points that have changed after matching the two images before and after treatment. The difference before and after treatment is determined by calculating the deformation coefficient of each point in the deformation field. The deformation coefficient is generally divided into three cases: deformation coefficient is 1, deformation coefficient>1 and deformation coefficient<1. For example, when the deformation coefficient is 1, it means that there is no change before and after treatment. If the local brain tissue corresponding to the deformation coefficient of 1 is abnormal brain tissue, it proves that there is no effect before and after treatment; if the local brain tissue corresponding to the deformation coefficient of 1 is non-abnormal brain tissue, it proves that the treatment process has no effect on the non-abnormal local brain tissue.

[0103] In this step, the corresponding deformation coefficient can be determined by each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map corresponding to each gestational age. After this step, the deformation coefficient determined according to each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map for all gestational ages can be obtained.

[0104] Step S150: For each local brain tissue, a linear analysis graph corresponding to each local brain tissue is generated according to the deformation coefficients of the local brain tissue corresponding to all gestational ages.

[0105] Specifically, the linear analysis graph corresponding to each local brain tissue is a two-dimensional image, in which the horizontal axis represents the gestational age and the vertical axis represents the deformation coefficient. The linear analysis graph corresponding to each local brain tissue contains multiple gestational ages and the deformation coefficient corresponding to each gestational age.

[0106] Step S160: determining final brain tissue analysis information based on the linear analysis graphs corresponding to all local brain tissues. The brain tissue analysis information is used to analyze the correlation between the treatment effect and the gestational age.

[0107] Specifically, the linear analysis graphs corresponding to the seven local brain tissues constitute the final brain tissue analysis information. The final brain tissue information can be in the form of text or an analysis graph. The specific form is not specifically limited in this embodiment of the application, as long as the data formed by the linear analysis graphs corresponding to the seven local brain tissues can be viewed through the final brain tissue information. The linear analysis graph corresponding to each local brain tissue allows medical personnel to understand the changes in each local tissue before and after treatment for a fetus with a congenital disease. Medical personnel can improve treatment methods based on the changes in each local tissue before and after treatment to enhance the treatment effect.

[0108] In an embodiment of the present application, multiple post-treatment brain MRI images of different gestational ages are obtained, and brain tissue segmentation is performed on each post-treatment brain MRI image to obtain multiple local brain tissues, and multiple post-treatment local brain tissue segmentation maps corresponding to the multiple local brain tissues are obtained. For each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue is obtained. Secondly, for each post-treatment brain MRI image, the deformation coefficient of each local brain tissue is determined based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, a linear analysis map corresponding to each local brain tissue is generated according to the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, final brain tissue analysis information is determined based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct an overall analysis of the impact of changes before and after treatment of different gestational ages and different brain tissues.

[0109] Furthermore, when brain tissue segmentation is performed on each post-treatment brain MRI image in step S120 using a trained pyramid segmentation model, the training process of the pyramid segmentation model includes steps S1 (not shown in the drawings), S2 (not shown in the drawings), S3 (not shown in the drawings), S4 (not shown in the drawings), and S5 (not shown in the drawings), wherein:

[0110] Step S1: Acquire multiple brain MRI training samples, each of which includes a brain MRI image and corresponding multiple local brain tissue labels.

[0111] Specifically, brain MRI training samples are obtained from a historical database of treatment cases. Local brain tissue labels corresponding to each training brain MRI image can be manually annotated. Multiple local brain tissue labels are labeled based on the local tissues in the brain MRI images, including cerebrospinal fluid, gray matter, white matter, ventricles, cerebellum, deep gray matter, and brainstem. Training samples containing local brain tissue labels facilitates determining the training level of the pyramid segmentation model based on the training results.

[0112] The form of multiple local brain tissue labels can be different labels for different segmented areas. The form of the labels is not specifically limited in the embodiment of the present application, as long as the local brain tissue labels can be used to distinguish the multiple local brain tissues of each brain MRI training sample.

[0113] Step S2: Obtain the pyramid segmentation model to be trained.

[0114] The pyramid segmentation model to be trained includes an encoder, a decoder, a skip connection module and a feature combination module.

[0115] Specifically, the pyramid segmentation model to be trained consists of a backbone network, which consists of four parts: an encoder, a decoder, a skip connection module, and a feature combination module. The encoder contains multiple encoding layers, and the encoding layers include convolution layers and pooling layers. The number of corresponding feature channels is increased by convolution and pooling the input feature map.

[0116] The decoder includes multiple decoding layers, each of which includes upsampling the input feature map to reduce the number of feature channels. Through the jump connection module, each decoding layer is connected in series with the features of the corresponding position in the encoder. The feature channels of the upsampled part allow the network to propagate contextual information to higher resolution layers, and finally output multiple target features. The multiple target features are combined through the feature combination module to obtain multiple local brain tissue segmentation maps.

[0117] Step S3: Input multiple brain MRI training samples into the pyramid segmentation model to be trained for iterative training to obtain a training result for each brain MRI training sample.

[0118] Specifically, the multiple brain MRI training samples include brain MRI images corresponding to healthy fetuses of different gestational ages, as well as brain MRI images corresponding to sick children of different gestational ages. The training sample training results are multiple local brain tissue segmentation maps obtained by segmenting the brain MRI training samples.

[0119] Step S4: Determine the function loss value based on each training result and the corresponding local brain tissue label.

[0120] Specifically, each brain MRI training sample corresponds to multiple local brain tissue labels. By comparing the training results with the corresponding local brain tissue labels, a corresponding function loss value can be obtained, where the function loss value represents the degree of difference between the training results and the corresponding local brain tissue labels. When determining the function loss value, a cross-entropy loss function or a logarithmic loss function can be used. The type of loss function is not specifically limited in the embodiments of this application, as long as the function loss value can be calculated through the loss function.

[0121] Step S5: If the function loss value is lower than the preset standard loss value, the iterative training is stopped to obtain a trained pyramid segmentation model.

[0122] Specifically, the preset standard loss value can be modified as needed and is not specifically limited in the embodiments of this application. The output training results are compared with the actual local brain tissue labels to obtain a function loss value. When the function loss value is within the preset loss value range, the pyramid segmentation model training is determined to be complete. The smaller the function loss value, the better the performance of the resulting pyramid segmentation model.

[0123] To facilitate understanding, an example is used for illustration. Figure 3 (a) is a brain MRI image during the pyramid model training process, Figure 3 (b) is a map of multiple local brain tissue labels corresponding to the brain MRI image during the pyramid segmentation model training process, and Figure 3 (c) is a map of the segmentation results of the pyramid segmentation model.

[0124] In an embodiment of the present application, multiple brain MRI training samples are obtained, wherein each brain MRI training sample includes a brain MRI image and corresponding multiple local brain tissue labels, and then the multiple brain MRI training samples are input into the obtained pyramid segmentation model to be trained for iterative training, and a training result corresponding to each brain MRI training sample is output, wherein the pyramid segmentation model to be trained includes an encoder, a decoder, a jump connection module and a feature combination module, and then a function loss value is determined by each training result and the corresponding local brain tissue label. When the function loss value is lower than a preset standard loss value, the iterative training of the pyramid segmentation model to be trained is stopped to obtain a trained pyramid segmentation model. Through iterative training, the accuracy of training the pyramid segmentation model is helped to be improved.

[0125] Furthermore, in step S120, brain tissue segmentation is performed on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues, including step S1201 (not shown in the drawings), step S1202 (not shown in the drawings), step S1203 (not shown in the drawings), and step S1204 (not shown in the drawings), wherein:

[0126] Step S1201: Acquire post-treatment brain MRI image features corresponding to each post-treatment brain MRI image.

[0127] Specifically, the brain MRI image after treatment can be a two-dimensional image or a three-dimensional image, where the two-dimensional image can be expressed as C*H*W, and the three-dimensional image can be expressed as C*H*W*D, where C represents the number of channels, H represents the image height, W represents the image width, and D represents the image depth.

[0128] Step S1202: Input each post-treatment brain MRI image feature into the encoder for feature extraction, and output the encoded MRI image feature corresponding to each encoding layer.

[0129] Among them, the feature extraction process is the encoding process, the encoder includes multiple encoding layers, and each encoding layer includes a downsampling module and an attention structure module.

[0130] Specifically, each downsampling module consists of convolutional and pooling layers, and the attention structure module is a COT-Block (Contextual Transformer block). Through several downsampling layers in the encoder, the feature channels of the treated brain MRI images are amplified, essentially shrinking the treated brain MRI images. The COT-Block attention structure model is a novel attention structure model that leverages contextual information of key values ​​to guide the learning of a dynamic attention matrix, thereby enhancing visual representation capabilities.

[0131] Each coding layer includes a downsampling module and an attention structure model. The attention structure model is used to realize the feature extraction and utilization of the brain MRI images after treatment. The brain MRI images after treatment will output a corresponding coded MRI image feature after passing through each coding layer. The coded MRI image feature output by the current coding layer serves as the input content of the next coding layer.

[0132] Step S1203: Based on the decoder and the skip connection module, the MRI image features corresponding to all decoding layers are feature-connected, and the target image features corresponding to each decoding layer are output.

[0133] The feature concatenation process is the decoding process. The decoder includes multiple decoding layers, each of which includes an upsampling module and an attention structure module. The attention structure model in the decoder can be the same as or different from the attention structure model included in the encoder. The last decoding layer also includes a hybrid dilated convolution module.

[0134] Specifically, each upsampling module contains a convolutional layer and a pooling layer, and the attention structure module is a COT-Block (Contextual Transformer block). Through several upsampling layers in the decoder, the feature channels of the post-treatment brain MRI image features are reduced, that is, the post-treatment brain MRI image is amplified.

[0135] The decoder input is the encoded MRI image features output by the last layer of the encoder, as well as the output results corresponding to the encoder position obtained through the jump connection module, and outputs multiple target image features.

[0136] The hybrid dilated convolution module is a HDC module. This module increases the receptive field of the convolution kernel, allowing it to output a wider range of information during the convolution process, thereby preserving a wider range of spatial information. For example, when performing local brain tissue segmentation on post-treatment brain MRI images, hybrid dilated convolution can expand boundary information and reduce the chance of edge information loss.

[0137] The encoder is combined with the decoder to increase or decrease the feature channels corresponding to the brain MRI images after treatment, so as to improve the accuracy of segmentation of the brain MRI images after treatment.

[0138] Step S1204: Input the multiple target image features into the feature combination module, and output multiple post-treatment local brain tissue segmentation maps.

[0139] Specifically, the local brain tissue segmentation map after treatment is the segmentation map corresponding to each local brain tissue. After obtaining multiple target image features, the multiple target features are fused to obtain an initial segmentation map. Then, the eigenvalue channels corresponding to the initial segmentation map are changed through convolution and pooling to obtain multiple local brain tissue segmentation maps after treatment.

[0140] In an embodiment of the present application, the MRI image features corresponding to each post-treatment brain MRI image are first obtained, and then each post-treatment brain MRI image feature is input into the encoder for feature extraction, and the encoded MRI image features corresponding to each encoding layer are output, wherein the encoder includes multiple encoding layers, and each encoding layer includes a downsampling module and an attention structure module. Then, based on the decoder and the jump connection module, all the encoded MRI images corresponding to the decoding layer are feature-connected to obtain multiple target image features. Finally, the multiple target image features are input into the feature combination module to obtain multiple post-treatment local brain tissue segmentation maps, thereby enhancing the accuracy of local brain tissue segmentation and thereby improving the segmentation accuracy.

[0141] Furthermore, in step S1202, each post-treatment brain MRI image feature is input into an encoder for feature extraction, and the encoded MRI image features corresponding to each encoding layer are output. This specifically includes step S1202a (not shown in the drawings) and step S1202b (not shown in the drawings), wherein:

[0142] Step S1202a: Input each post-treatment brain MRI image feature into the first encoding layer, and output the first encoded MRI image feature.

[0143] Step S1202b: Input the first encoded MRI image features into the second encoding layer to obtain the second encoded MRI image features, and execute the loop until the last encoding layer in the encoder outputs the last layer of encoded MRI image features to obtain the encoded MRI image features corresponding to each encoding layer.

[0144] For easier understanding, please refer to Figure 4 , Figure 4This is a structural diagram of the pyramid segmentation model in an embodiment of the present application. The encoder includes multiple encoding layers, each of which includes a downsampling module and an attention structure module COT-Block. The downsampling module is used to convolve and pool the input brain MRI image to change the size of the feature channel corresponding to the post-treatment brain MRI image feature input to the first encoding layer. The first encoded MRI image feature is then output through the attention structure model COT-Block, and the output first encoded MRI image is input to the next encoding layer to continue outputting the second encoded MRI image. The content output by the current encoding layer is the content input by the next encoding layer. Each encoding layer corresponds to an output result, which is the encoded MRI image feature corresponding to each encoding layer.

[0145] In an embodiment of the present application, by inputting each post-treatment brain MRI image feature into the first coding layer, outputting the first coded MRI image feature, and then inputting the first coded MRI image feature into the second coding layer to obtain the second coded MRI image feature, the above steps are repeated until the last coding layer in the encoder outputs the last layer of coded MRI image features, and finally obtaining the coded MRI image features corresponding to each coding layer, and adding an attention structure model to the encoder to improve the accuracy of extracting the coded MRI image features.

[0146] Furthermore, in step S1203, based on the decoder and the skip connection module, the MRI image features corresponding to all decoding layers are feature-connected, and the target image features corresponding to each decoding layer are output. Specifically, the step S1203a (not shown in the drawings) and step S1203b (not shown in the drawings) are included, wherein:

[0147] Step S1203a: Input the encoded MRI image features output by the last layer in the encoder into the first decoding layer to obtain first decoded MRI image features.

[0148] Step S1203b: Based on the skip connection module, the first decoded MRI image features and the encoded MRI image features of the corresponding encoding layer are input into the second decoding layer to obtain the second decoded MRI image features, and the loop is executed until the last decoding layer in the decoder to obtain the target image features corresponding to each decoding layer.

[0149] For details, please continue to refer to Figure 4 , Figure 4 This is a structural diagram of the pyramid segmentation model in an embodiment of the present application, encoder part: after the post-treatment brain MRI image features are input into the first encoding layer, the first encoding layer will output the first encoded MRI image features, and the first encoded MRI image features are input into the second encoding layer, the second encoding layer outputs the second encoded MRI image features, until the fourth encoding layer outputs the fourth encoded MRI image.

[0150] During the encoding process, when the input parameters of each encoding layer are updated, the corresponding output parameters will also be updated accordingly. This is called internal covariate shift. The more layers there are in the encoding process, the more obvious this phenomenon is. Internal covariate shift will cause many problems in the encoding process:

[0151] (1) Each encoding layer needs to constantly adapt to changes in input content, which may affect encoding efficiency and make the encoding process unstable;

[0152] (2) The update of the input content of the first few layers in the encoding process may cause the input content of the later layers to become too large or too small, thus falling into the saturation region of the activation function, causing the encoding process to stop prematurely.

[0153] Therefore, in the embodiment of the present application, the encoding process also includes a batch normalization processing module to facilitate enhancing the stability of the encoding process, and the corresponding decoding process also includes a batch normalization processing module to facilitate enhancing the stability of the decoding process.

[0154] Decoder part: Input the fourth coded MRI image output by the fourth coding layer into the first decoding layer, and then obtain the fourth coded MRI image features corresponding to the fourth coding layer through the jump connection module, and splice the two to obtain the first decoded MRI image features, input the spliced ​​first spliced ​​image features into the second decoding layer, and output the first target image features; in the second decoding layer, splice the input first spliced ​​image features with the third coded MRI image features corresponding to the third coding layer to obtain the second decoded image features, input the spliced ​​second spliced ​​image features into the third decoding layer, and output the second target image features, until the fourth decoding layer outputs the fourth target features. Figure 4 In the example shown, the encoder has 5 layers and the decoder has 4 layers, the second encoding layer of the encoder corresponds to the fourth decoding layer of the decoder, the third encoding layer of the encoder corresponds to the third decoding layer of the decoder, and the fourth encoding layer of the encoder corresponds to the first decoding layer of the decoder. Figure 4 In the , copy and crop represent the process of skip connection operation, conv3d and up_sampling represent the process of three-dimensional convolution and up-sampling operation.

[0155] In an embodiment of the present application, the encoded MRI image features output by the last layer in the encoder are input into the first decoding layer to obtain the first decoded MRI image features, and then based on the jump connection module, the encoded MRI image features of the corresponding encoding layer are obtained to complete the splicing, and the splicing result is input into the second decoding layer to obtain the second decoded MRI image features and the corresponding target image features, until the last decoding layer in the decoder outputs the corresponding target image, so as to obtain the target image features corresponding to each decoding layer. The combination of the encoding process and the decoding process is realized through the jump link module, which helps to improve the accuracy of obtaining the target image features.

[0156] Furthermore, the attention structure module is used for: step Sa (not shown in the drawings), step Sb (not shown in the drawings), step Sc (not shown in the drawings), step Sd (not shown in the drawings), step Se (not shown in the drawings), and step Sf (not shown in the drawings), wherein:

[0157] Step Sa: Obtain the pixel feature values ​​of the sampled image. When the attention structure module is the attention structure module of the encoding layer, the sampled image is the downsampled brain MRI image after treatment. When the attention structure module is the attention structure module of the decoding layer, the sampled image is the upsampled encoded MRI image corresponding to the decoding layer.

[0158] Specifically, the pixel characteristic value corresponding to the three-dimensional brain MRI image after treatment is C*H*W*D, where C represents the number of channels of the three-dimensional brain MRI image after treatment, H represents the image height of the three-dimensional brain MRI image after treatment, W represents the image width of the three-dimensional brain MRI image after treatment, and D represents the image depth of the three-dimensional brain MRI image after treatment.

[0159] The pixel characteristic value corresponding to the two-dimensional brain MRI image after treatment is C*H*W, where C represents the number of channels of the two-dimensional brain MRI image after treatment, H represents the image height of the two-dimensional brain MRI image after treatment, and W represents the image width of the two-dimensional brain MRI image after treatment.

[0160] Step Sb: Determine the corresponding matrix Q, matrix K and matrix V according to the pixel eigenvalues.

[0161] For details, please refer to Figure 5 , Figure 5 This is a schematic diagram of the structure of the attention structure module in the pyramid segmentation model implemented in this application. If it is an attention structure module in the encoder, the image eigenvalues ​​corresponding to the sampled post-treatment brain MRI image are input into the matrix determination model to obtain the corresponding matrices Q, K, and V, where matrix Q is the query matrix, matrix K is the key matrix, and matrix V is the value matrix, and the matrix determination model can be an embedding matrix.

[0162] Step Sc: Convolve the matrix K to obtain the matrix K1.

[0163] Specifically, the matrix K is convolved with a k*k*k group to obtain the matrix K1 to establish a context connection. If the brain MRI image after treatment is a two-dimensional image, a k*k group convolution is performed on it. The value of k can be modified according to user needs and is not specifically limited in the embodiment of this application.

[0164] Step Sd: Concatenate the matrix K1 and the matrix Q to obtain an initial matrix, and perform two consecutive convolution operations on the initial matrix to obtain a relational matrix A.

[0165] Specifically, the matrix is ​​connected, that is, two different matrices are spliced ​​together, and the initial matrix obtained after the matrix K1 and the matrix Q are connected is [K1, Q]. After two consecutive 1*1*1 convolutions on the initial matrix, the relationship matrix A is obtained, and the relationship matrix A=[K1, Q]W θ W δ , which is convenient for enriching the relationship mapping of context feature information, where W θ Represents a 1*1*1 convolution operation, W δ Represents a 1*1*1 convolution operation.

[0166] Step Se: Perform matrix multiplication on the relationship matrix A and V to obtain the matrix K2.

[0167] Step Sf: Fuse matrix K1 and matrix K2 to output image features, where when the attention structure module is the attention structure module of the encoding layer, the encoded MRI image features are output; when the attention structure module is the attention structure module of the decoding layer, the target image features are output.

[0168] Specifically, after matrix A and matrix V are further multiplied, K2=[A*V] is obtained. Matrix multiplication facilitates capturing the corresponding dynamic feature interactions.

[0169] In an embodiment of the present application, the attention structure model is applied in both the encoder and the decoder. When the attention structure module is the attention structure module of the encoding layer, the sampled image is the upsampled brain MRI image after treatment. When the attention structure module is the attention structure module of the decoding layer, the sampled image is the downsampled encoded MRI image corresponding to the decoding layer. After obtaining the corresponding image feature values, the corresponding matrix Q, matrix K and matrix V are obtained. By performing correlated convolution, splicing, multiplication and fusion on the matrix Q, matrix K and matrix V, the image features are finally output. When the attention structure module is the attention structure module of the decoding layer, the sampled image is the downsampled encoded MRI image corresponding to the decoding layer. By establishing contextual connections, the accuracy of outputting image features is improved.

[0170] Furthermore, in step S130, for each post-treatment brain MRI image, obtaining a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map may specifically include step S1301 (not shown in the drawings), step S1302 (not shown in the drawings), step S1303 (not shown in the drawings), and step S1304 (not shown in the drawings), wherein:

[0171] Step S1301: Determine the gestational age based on each post-treatment brain MRI image.

[0172] Step S1302: According to each post-treatment brain MRI image and the corresponding gestational age, the corresponding pre-treatment brain MRI image is determined from the historical database.

[0173] Specifically, each post-treatment brain MRI image corresponds to a different gestational age, and the gestational age can be determined according to a preset rule. The preset rule is determined by the size and morphology of each local brain tissue in the post-treatment brain MRI image corresponding to different gestational ages in the historical treatment data information.

[0174] After the gestational age is determined, the corresponding pre-treatment brain MRI image is determined by traversing the historical database according to the post-treatment brain MRI image and the gestational age, wherein the historical database stores all pre-treatment brain MRI images.

[0175] Step S1303: Determine a plurality of corresponding pre-treatment local brain tissue segmentation maps based on each pre-treatment brain MRI image.

[0176] Step S1304: for each local brain tissue, determining the local brain tissue segmentation map before treatment corresponding to each local brain tissue segmentation map after treatment.

[0177] Specifically, the method for obtaining multiple pre-treatment local brain tissue segmentation maps based on the pre-treatment brain MRI image is the same as the method for obtaining multiple post-treatment local brain tissue segmentation maps by segmenting the post-treatment brain MRI image. The pre-treatment local brain tissue segmentation maps can be segmented in advance and stored in the same location as the pre-treatment brain MRI images in the historical database, or they can be segmented after traversing the historical database.

[0178] For easier understanding, please refer to Figure 6 , Figure 6This is a flowchart for local brain tissue analysis. After local brain tissue segmentation is performed on the brain MRI images before and after treatment, the local brain tissue segmentation maps before and after treatment are matched according to the name of each local brain tissue. The matching can also be performed by the edge contour of each local brain tissue. The specific matching method is not specifically limited in the embodiments of this application. As long as the local brain tissue segmentation maps before and after treatment can be matched, the corresponding results are determined by morphometrics (Deformation-Based Morphometry, DBM). That is, the deformation field formed by each local brain tissue before and after treatment is determined, the logarithm of the Jacobian determinant of the deformation field of each local brain tissue is calculated, and then all local brain tissue coefficients are uniformly analyzed to facilitate medical personnel to study and improve treatment methods based on changes in local brain tissue.

[0179] In an embodiment of the present application, the corresponding gestational age is determined based on the post-treatment brain MRI image, and the corresponding pre-treatment brain MRI image is matched from the historical database based on the gestational age and the post-treatment brain MRI image. Then, based on the local brain tissue, the local brain tissue segmentation maps corresponding to before and after treatment are matched. By traversing the historical database, the accuracy of determining the corresponding local brain tissue segmentation map before treatment is improved.

[0180] Furthermore, in step S140, for each post-treatment brain MRI image, the deformation coefficient of each local brain tissue is determined based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Specifically, the step S1401 (not shown in the drawings), step S1402 (not shown in the drawings), step S1403 (not shown in the drawings), and step S1404 (not shown in the drawings) are included, wherein:

[0181] Step S1401: For each post-treatment brain MRI image, each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map are imported into a pre-established coordinate system to determine the pre- and post-treatment boundary point coordinates of each pre-treatment local brain tissue segmentation map and each corresponding post-treatment local brain tissue segmentation map.

[0182] Specifically, the coordinates of the boundary points before and after treatment include the coordinates of all points on the contour edge of the local brain tissue segmentation map before treatment and the coordinates of all points on the contour edge of the local brain tissue segmentation map after treatment. The contour edge in each local brain tissue segmentation map can be identified using an image recognition model.

[0183] Step S1402: For each local brain tissue, determine the coordinates of the change point set based on the coordinates of all boundary points before and after treatment of each local brain tissue.

[0184] Specifically, taking the ventricles of local brain tissue as an example, when determining change points based on the coordinates of the corresponding boundary points in the pre-treatment and post-treatment ventricle segmentation maps, the coordinates of the boundary points in the pre-treatment ventricle segmentation map can be used as target points, and the boundary points in the post-treatment ventricle segmentation map can be used as matching points. The coordinates of each matching point are determined to be identical to the target point coordinates at the corresponding position. If they are different, the matching point is determined as a change point, and the coordinates of the matching point are determined as the change point coordinates. The change point set coordinates contain multiple change point coordinates and corresponding target point coordinates.

[0185] Step S1403: Determine the deformation field of each brain tissue according to the coordinates of the change point set corresponding to each local brain tissue.

[0186] Specifically, the brain tissue deformation field is a change field composed of multiple change points and target point movements.

[0187] Step S1404: Determine the deformation coefficient corresponding to the deformation field of each local brain tissue.

[0188] Specifically, the deformation coefficient corresponding to each local brain tissue deformation field is the average value of the initial deformation coefficients of all change point sets. The specific calculation process of the initial deformation coefficient corresponding to any change point is as follows:

[0189] For example, if the coordinates of any point in the deformation field are p(a, b, c), then the corresponding Jacobian is

[0190] The logarithm of the formula, that is, the deformation coefficient, is:

[0191] det(p(a,b,c))

[0192] Here, "det" is a qualifier for taking the differential of any point in the deformation field. A deformation coefficient of 1 indicates no change before and after treatment. A deformation coefficient less than 1 indicates that the local brain tissue has shrunk after treatment. A deformation coefficient greater than 1 indicates that the local brain tissue has expanded after treatment. The treatment effect is related to the deformation coefficient. For example, if a fetus suffers from the congenital disease spina bifida, the ventricles of the child will be larger and longer than those of a normal fetus. A deformation coefficient of 1 after treatment indicates no treatment effect. A deformation coefficient less than 1 indicates a good treatment effect. A deformation coefficient greater than 1 indicates worsening of the condition.

[0193] In the embodiment of the present application, each local brain tissue segmentation map before treatment and the corresponding local brain tissue segmentation map after treatment are imported into the established coordinate system, the boundary point coordinates are determined respectively, and then the coordinates of the change point set before and after treatment corresponding to the local brain tissue are determined by the boundary point coordinates, and then the corresponding deformation field is determined, and finally the deformation coefficient corresponding to the deformation field of each local brain tissue is calculated. The deformation coefficient facilitates a comprehensive analysis of the impact of the treatment effect on each local brain tissue before and after treatment.

[0194] The above embodiment introduces a method for fetal brain MRI tissue analysis from the perspective of method flow. The following embodiment introduces a device for fetal brain MRI tissue analysis from the perspective of virtual modules or virtual units. Please refer to the following embodiments for details.

[0195] The present application embodiment provides a device for fetal brain MRI tissue analysis, such as Figure 7 As shown, the apparatus may specifically include an image acquisition module 710, a brain tissue segmentation module 720, a corresponding image acquisition module 730, a deformation coefficient determination module 740, an analysis graph generation module 750, and an analysis information determination module 760, wherein:

[0196] The image acquisition module 710 is configured to acquire a plurality of post-treatment brain MRI images.

[0197] Among them, each post-treatment brain MRI image corresponds to a different gestational age;

[0198] A brain tissue segmentation module 720 is configured to perform brain tissue segmentation on each post-treatment brain MRI image to obtain a plurality of post-treatment local brain tissue segmentation maps corresponding to a plurality of local brain tissues;

[0199] The corresponding image acquisition module 730 is configured to acquire, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map;

[0200] a deformation coefficient determination module 740 for determining, for each post-treatment brain MRI image, a deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map;

[0201] An analysis graph generating module 750 is configured to generate a linear analysis graph corresponding to each local brain tissue according to the deformation coefficients of the local brain tissue corresponding to all fetal ages;

[0202] The analysis information determination module 760 is configured to determine final brain tissue analysis information based on the linear analysis graphs corresponding to all local brain tissues.

[0203] The final brain tissue analysis information was used to analyze the association between treatment effect and gestational age.

[0204] In one possible implementation, the training process of the pyramid segmentation model in the brain tissue segmentation module 720 includes:

[0205] Acquire a training sample unit, obtain multiple brain MRI training samples, each brain MRI training sample includes a brain MRI image and corresponding multiple local brain tissue labels;

[0206] Get the model to be trained unit, used to obtain the pyramid segmentation model to be trained.

[0207] The pyramid segmentation model to be trained includes an encoder, a decoder, a skip connection module and a feature combination module;

[0208] an iterative training unit, configured to input a plurality of brain MRI training samples into the pyramid segmentation model to be trained for iterative training, and obtain a training result for each brain MRI training sample;

[0209] a function loss value determining unit, configured to determine a function loss value based on each training result and a corresponding local brain tissue label;

[0210] The judgment unit is used to stop iterative training if the function loss value is lower than a preset standard loss value, thereby obtaining a trained pyramid segmentation model.

[0211] In one possible implementation, the brain tissue segmentation module 720 includes:

[0212] a feature acquisition unit, configured to acquire a post-treatment brain MRI image feature corresponding to each post-treatment brain MRI image;

[0213] The encoding MRI image feature acquisition unit is used to input each post-treatment brain MRI image feature into the encoder for feature extraction, and output the encoding MRI image feature corresponding to each encoding layer.

[0214] The encoder includes multiple encoding layers, each of which includes a downsampling module and an attention structure module;

[0215] The target image feature acquisition unit is used to perform feature connection on the MRI image features corresponding to all decoding layers based on the decoder and the jump connection module, and output the target image features corresponding to each decoding layer.

[0216] The decoder includes multiple decoding layers, each of which includes an upsampling module and an attention structure module, and the last decoding layer also includes a hybrid dilated convolution module;

[0217] The local brain tissue segmentation map output unit is used to input multiple target image features into the feature combination module for feature combination, and output multiple local brain tissue segmentation maps after treatment.

[0218] In one possible implementation, obtaining the encoded MRI image feature unit includes:

[0219] a first encoding execution subunit, configured to input each post-treatment brain MRI image feature into a first encoding layer and output a first encoded MRI image feature;

[0220] The second encoding execution subunit is used to input the first encoding MRI image feature into the second encoding layer to obtain the second encoding MRI image feature, and execute the loop until the last encoding layer in the encoder outputs the last layer of encoding MRI image features to obtain the encoding MRI image features corresponding to each encoding layer.

[0221] In one possible implementation, obtaining a target image feature unit includes:

[0222] a first decoding execution subunit, configured to input the encoded MRI image features output by the last layer of the encoder into the first decoding layer to obtain first decoded MRI image features;

[0223] The second decoding execution subunit is used to input the first decoded MRI image features and the encoded MRI image features of the corresponding encoding layer into the second decoding layer based on the skip connection module to obtain the second decoded MRI image features, and execute the loop until the last decoding layer in the decoder to obtain the target image features corresponding to each decoding layer.

[0224] In one possible implementation, in the unit for obtaining the encoded MRI image features, the process of outputting the encoded MRI image features through the attention structure module includes:

[0225] Determine a pixel feature value subunit, which is used to obtain pixel feature values ​​of the sampled image, when the attention structure module is the attention structure module of the encoding layer, the sampled image is a downsampled post-treatment brain MRI image; when the attention structure module is the attention structure module of the decoding layer, the sampled image is an upsampled encoded MRI image corresponding to the decoding layer;

[0226] Obtain a matrix subunit, for determining corresponding matrices Q, K, and V according to pixel eigenvalues;

[0227] The matrix convolution subunit is used to convolve the matrix K to obtain the matrix K1;

[0228] The matrix connection subunit is used to connect the matrix K1 and the matrix Q to obtain the initial matrix, and perform two consecutive convolution operations on the initial matrix to obtain the relationship matrix A;

[0229] A matrix multiplication subunit, used for performing matrix multiplication on the relation matrix A and the matrix V to obtain the matrix K2;

[0230] The matrix fusion subunit is used to fuse the matrix K1 and the matrix K2 to output image features, wherein when the attention structure module is the attention structure module of the encoding layer, the encoded MRI image features are output; when the attention structure module is the attention structure module of the decoding layer, the target image features are output.

[0231] In one possible implementation, the corresponding image acquisition module 630 includes:

[0232] a gestational age determination unit, configured to determine the gestational age based on each post-treatment brain MRI image;

[0233] a pre-treatment image determination unit, configured to determine a corresponding pre-treatment brain MRI image from a historical database according to each post-treatment brain MRI image and the corresponding gestational age;

[0234] a pre-treatment segmentation map determining unit, configured to determine a plurality of corresponding pre-treatment local brain tissue segmentation maps based on each pre-treatment brain MRI image;

[0235] The pre- and post-treatment segmentation map corresponding unit is used to determine, for each local brain tissue, the pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map.

[0236] In one possible implementation, the deformation coefficient determination module 740 includes:

[0237] a boundary point coordinate determination unit, configured to import each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map into a pre-established coordinate system for each post-treatment brain MRI image, and determine the pre- and post-treatment boundary point coordinates of each pre-treatment local brain tissue segmentation map and each corresponding post-treatment local brain tissue segmentation map;

[0238] a change point set determination unit, configured to determine the coordinates of the change point set for each local brain tissue according to the coordinates of all boundary points before and after treatment of each local brain tissue;

[0239] a deformation field determination unit, configured to determine the deformation field of each brain tissue according to the coordinates of the change point set corresponding to each local brain tissue;

[0240] The deformation coefficient determining unit is used to determine the corresponding deformation coefficient of each local brain tissue according to each brain tissue deformation field.

[0241] An electronic device is provided in an embodiment of the present application, such as Figure 8 As shown, Figure 8The electronic device 800 shown includes a processor 801 and a memory 803. The processor 801 and the memory 803 are connected, for example, via a bus 802. Optionally, the electronic device 800 may further include a transceiver 804. It should be noted that in actual applications, the number of transceivers 804 is not limited to one, and the structure of the electronic device 800 does not constitute a limitation on the embodiments of the present application.

[0242] Processor 801 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 801 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0243] Bus 802 may include a path for transmitting information between the aforementioned components. Bus 802 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. Bus 802 may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0244] The memory 803 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0245] The memory 803 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 801. The processor 801 is used to execute the application code stored in the memory 803 to implement the content shown in the above method embodiment.

[0246] Electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. They may also include servers, etc. Figure 8 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0247] The present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed on a computer, the computer can execute the corresponding contents of the aforementioned method embodiment. Compared with the related art, the present invention obtains a plurality of brain MRI images after treatment at different gestational ages, performs brain tissue segmentation on each brain MRI image after treatment to obtain a plurality of local brain tissues, and obtains a plurality of local brain tissue segmentation maps after treatment corresponding to the plurality of local brain tissues. For each brain MRI image after treatment, obtains a pre-treatment local brain tissue segmentation map corresponding to each local brain tissue after treatment. Then, for each brain MRI image after treatment, determine the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map. Then, for each local brain tissue, generate a linear analysis map corresponding to each local brain tissue based on the local brain tissue deformation coefficients corresponding to all gestational ages. Finally, determine final brain tissue analysis information based on the linear analysis maps corresponding to all local brain tissues. The final brain tissue analysis information facilitates relevant personnel to conduct a comprehensive analysis of the impact of changes before and after treatment on different gestational ages and different brain tissues.

[0248] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0249] The above description is only part of the implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A fetal brain MRI tissue analysis method, characterized in that: include: Acquiring a plurality of post-treatment brain MRI images, wherein each post-treatment brain MRI image corresponds to a different gestational age; Perform brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues; For each post-treatment brain MRI image, obtaining a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map; For each post-treatment brain MRI image, determining the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map; For each local brain tissue, a linear analysis graph corresponding to each local brain tissue is generated according to the local brain tissue deformation coefficient corresponding to all fetal ages; determining final brain tissue analysis information based on linear analysis graphs corresponding to all local brain tissues, wherein the brain tissue analysis information is used to analyze the correlation between treatment effect and fetal age; The step of performing brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues includes: Obtaining post-treatment brain MRI image features corresponding to each post-treatment brain MRI image; Input each post-treatment brain MRI image feature into an encoder for feature extraction, and output the encoded MRI image feature corresponding to each encoding layer, wherein the encoder includes multiple encoding layers, each encoding layer includes a downsampling module and an attention structure module; Based on the decoder and skip connection module, all the encoded MRI image features corresponding to the decoding layer are feature-connected, and the target image features corresponding to each decoding layer are output. The decoder includes multiple decoding layers, each decoding layer includes an upsampling module and an attention structure module, and the last decoding layer also includes a hybrid dilated convolution module; Input multiple target image features into the feature combination module and output multiple local brain tissue segmentation maps after treatment; The step of determining the deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map for each post-treatment brain MRI image includes: For each post-treatment brain MRI image, each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map are imported into a pre-established coordinate system to determine the coordinates of the pre- and post-treatment boundary points of each pre- and post-treatment local brain tissue segmentation map; For each local brain tissue, the coordinates of the change point set are determined based on the coordinates of all boundary points of each local brain tissue before and after treatment; Determine the deformation field of each brain tissue according to the coordinates of the change point set corresponding to each local brain tissue; The corresponding deformation coefficient of each local brain tissue is determined according to each brain tissue deformation field.

2. The fetal brain MRI tissue analysis method according to claim 1, characterized in that: The brain tissue segmentation of each post-treatment brain MRI image is achieved by a trained pyramid segmentation model. The training process of the pyramid segmentation model includes: Acquire multiple brain MRI training samples, each brain MRI training sample including a brain MRI image and corresponding multiple local brain tissue labels; Obtaining a pyramid segmentation model to be trained, wherein the pyramid segmentation model to be trained includes an encoder, a decoder, a skip connection module, and a feature combination module; Inputting multiple brain MRI training samples into the pyramid segmentation model to be trained for iterative training to obtain a training result for each brain MRI training sample; Determine the function loss value based on each training result and the corresponding local brain tissue label; If the function loss value is lower than the preset standard loss value, the iterative training is stopped to obtain a trained pyramid segmentation model.

3. The fetal brain MRI tissue analysis method according to claim 1, characterized in that: The step of inputting each post-treatment brain MRI image feature into an encoder for feature extraction and outputting the encoded MRI image feature corresponding to each encoding layer includes: Input each post-treatment brain MRI image feature into a first encoding layer, and output a first encoded MRI image feature; The first encoded MRI image feature is input into the second encoding layer to obtain the second encoded MRI image feature, and the loop is executed until the last encoding layer in the encoder outputs the last layer of encoded MRI image features to obtain the encoded MRI image features corresponding to each encoding layer.

4. The fetal brain MRI tissue analysis method according to claim 1, characterized in that: The decoder and skip connection module performs feature connection on all encoded MRI image features corresponding to the decoding layer and outputs the target image features corresponding to each decoding layer, including: Inputting the encoded MRI image features output by the last layer in the encoder into the first decoding layer to obtain first decoded MRI image features; Based on the jump connection module, the first decoded MRI image features and the encoded MRI image features of the corresponding encoding layer are input into the second decoding layer to obtain the second decoded MRI image features. The loop is executed until the last decoding layer in the decoder to obtain the target image features corresponding to each decoding layer.

5. The fetal brain MRI tissue analysis method according to claim 1, characterized in that: The attention structure module is used to: Obtaining pixel feature values ​​of the sampled image, when the attention structure module is the attention structure module of the encoding layer, the sampled image is a downsampled brain MRI image after treatment; when the attention structure module is the attention structure module of the decoding layer, the sampled image is an upsampled encoded MRI image corresponding to the decoding layer; Determine the corresponding matrix Q, matrix K and matrix V according to the pixel eigenvalue; Convolving the matrix K to obtain a matrix K1; Concatenate the matrix K1 and the matrix Q to obtain an initial matrix, and perform two consecutive convolution operations on the initial matrix to obtain a relational matrix A; Perform matrix multiplication on the relationship matrix A and the matrix V to obtain a matrix K2; The matrix K1 and the matrix K2 are fused to output image features, wherein when the attention structure module is the attention structure module of the encoding layer, the encoded MRI image features are output; when the attention structure module is the attention structure module of the decoding layer, the target image features are output.

6. The fetal brain MRI tissue analysis method according to claim 1, characterized in that: The step of obtaining, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map comprises: The gestational age was determined based on the brain MRI images after each treatment; According to each post-treatment brain MRI image and the corresponding gestational age, the corresponding pre-treatment brain MRI image is determined from the historical database; Determining a plurality of corresponding pre-treatment local brain tissue segmentation maps according to each pre-treatment brain MRI image; For each local brain tissue, a local brain tissue segmentation map before treatment corresponding to each local brain tissue segmentation map after treatment is determined.

7. A fetal brain MRI tissue analysis device, characterized in that: include: An image acquisition module is used to acquire a plurality of brain MRI images after treatment, wherein each brain MRI image after treatment corresponds to a different gestational age; A brain tissue segmentation module is used to perform brain tissue segmentation on each post-treatment brain MRI image to obtain multiple post-treatment local brain tissue segmentation maps corresponding to multiple local brain tissues; A corresponding image acquisition module is used to acquire, for each post-treatment brain MRI image, a pre-treatment local brain tissue segmentation map corresponding to each post-treatment local brain tissue segmentation map; a deformation coefficient determination module, configured to determine, for each post-treatment brain MRI image, a deformation coefficient of each local brain tissue based on each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map; An analysis graph generation module is used to generate, for each local brain tissue, a linear analysis graph corresponding to the local brain tissue according to the deformation coefficient of the local brain tissue corresponding to all fetal ages; an analysis information determination module, configured to determine final brain tissue analysis information based on linear analysis graphs corresponding to all local brain tissues, wherein the final brain tissue analysis information is used to analyze the correlation between treatment effect and gestational age; Among them, the brain tissue segmentation module includes: a feature acquisition unit, configured to acquire a post-treatment brain MRI image feature corresponding to each post-treatment brain MRI image; A coding MRI image feature acquisition unit is configured to input each post-treatment brain MRI image feature into an encoder for feature extraction, and output the coding MRI image feature corresponding to each coding layer, wherein the encoder includes multiple coding layers, each coding layer includes a downsampling module and an attention structure module; A target image feature acquisition unit is configured to perform feature connection on all encoded MRI image features corresponding to the decoding layer based on a decoder and a skip connection module, and output the target image features corresponding to each decoding layer. The decoder includes multiple decoding layers, each decoding layer includes an upsampling module and an attention structure module, and the last decoding layer also includes a hybrid dilated convolution module. The local brain tissue segmentation map output unit is used to input multiple target image features into the feature combination module and output multiple local brain tissue segmentation maps after treatment; The module for determining the deformation coefficient includes: a boundary point coordinate determination unit, configured to import each pre-treatment local brain tissue segmentation map and the corresponding post-treatment local brain tissue segmentation map into a pre-established coordinate system for each post-treatment brain MRI image, and determine the pre- and post-treatment boundary point coordinates of each pre-treatment local brain tissue segmentation map and each corresponding post-treatment local brain tissue segmentation map; a change point set determination unit, configured to determine the coordinates of the change point set for each local brain tissue according to the coordinates of all boundary points before and after treatment of each local brain tissue; a deformation field determination unit, configured to determine the deformation field of each brain tissue according to the coordinates of the change point set corresponding to each local brain tissue; The deformation coefficient determining unit is used to determine the corresponding deformation coefficient of each local brain tissue according to each brain tissue deformation field.

8. An electronic device, characterized in that: The electronic device includes: at least one processor; Memory; At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute the method for fetal brain MRI tissue analysis according to any one of claims 1-6.

Images