A method and system for assessing the state of fetal brain development

Abstract

This invention relates to a method and system for assessing fetal brain development, belonging to the field of brain development assessment technology. It solves the problems of strong subjectivity, low efficiency, and limited analytical dimensions in existing technologies for assessing fetal brain development. The method includes acquiring MRI images of the fetus to be assessed, performing preprocessing to obtain an initial fetal MRI image; obtaining fetal brain region images based on the initial fetal MRI image and a trained brain region recognition model; determining the gestational age stage of the fetal brain based on the fetal brain region images and a trained gestational age discriminator; obtaining segmented fetal brain region images based on the fetal brain region images, the gestational age stage of the fetal brain, and a trained segmented brain region model, and then extracting features from each fetal brain region; obtaining the equivalent gestational age of fetal brain development based on the features of each fetal brain region; further determining the developmental offset based on the actual gestational age of the fetus; and finally obtaining the fetal brain development assessment result.

Description

Technical Field

[0001] This invention relates to the field of brain development assessment technology, and in particular to a method and system for assessing fetal brain development status. Background Technology

[0002] With the rapid development of perinatal medicine and imaging technology, accurate prenatal diagnosis is of great significance for eugenics and improving population quality. As the most complex organ in the human body, the fetal brain's development directly affects the newborn's long-term neurological and cognitive functions. Magnetic resonance imaging (MRI), due to its high soft tissue resolution, has become a key technology besides ultrasound for assessing fetal brain development, especially for detecting subtle structural abnormalities. Quantitative analysis of brain region volume, morphology, and microstructure through MRI images is the core approach to achieving objective and accurate assessment of fetal brain development.

[0003] Currently, fetal brain MRI assessment mainly relies on two methods: one is visual observation and subjective judgment by radiologists based on clinical experience, and the other is simple manual or semi-automatic measurement using traditional image processing tools. However, subjective assessment is highly dependent on the doctor's personal experience. The results of manual measurement by the doctor are easily affected by factors such as personal experience and fatigue. There may be significant differences between different observers, and even between the same observer at different times. This has inherent drawbacks such as low efficiency, poor repeatability, and difficulty in quantification. Manual delineation and measurement are time-consuming and laborious, making them unsuitable for large-scale clinical screening or research. Furthermore, they cannot achieve comprehensive quantification of dozens of fine brain regions (such as the hippocampus, basal ganglia, and gray matter in specific functional areas), resulting in low efficiency and difficulty in scaling. In addition, existing methods mainly focus on macroscopic morphological parameters (such as volume) and cannot effectively extract and utilize the deep information (such as texture features) contained in the pixel intensity and spatial distribution of images. This microstructural information is crucial for early and mild developmental abnormalities.

[0004] In summary, existing technologies for assessing fetal brain development suffer from drawbacks such as high subjectivity, low efficiency, and limited analytical dimensions. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a method and system for assessing fetal brain development status, in order to solve the problems of strong subjectivity, low efficiency and single analysis dimension in the existing assessment of fetal brain development.

[0006] On one hand, embodiments of the present invention provide a method for assessing fetal brain development status, comprising the following steps: Acquire fetal MRI images to be evaluated, perform preprocessing, and obtain initial fetal MRI images; Based on the initial fetal MRI image and the trained brain region recognition model, a fetal brain region image is obtained; Based on the fetal brain region images and the trained gestational age discriminator, the gestational age stage of the fetal brain is obtained; Based on the fetal brain region images, the gestational age of the fetal brain, and the trained segmented brain region segmentation model, a segmented image of the fetal brain region is obtained, and then the features of each fetal brain region are extracted. Based on the characteristics of each brain region of the fetus, the equivalent gestational age of fetal brain development is obtained, and then the developmental deviation is obtained according to the actual gestational age of the fetus, thus obtaining the fetal brain development assessment results.

[0007] Furthermore, images of fetal brain regions were obtained through the following methods: The initial fetal MRI image is input into a trained brain region recognition model to obtain a binary mask of the fetal brain region; wherein, the brain region recognition model is YOLOv8-seg; The initial fetal MRI image is masked based on the binary mask of the fetal brain region to obtain an image of the fetal brain region.

[0008] Furthermore, the gestational age stages of the fetal brain include early, mid, and late stages; the segmented brain region segmentation model includes an early brain region segmentation model, a mid-term brain region segmentation model, and a late-term brain region segmentation model; wherein, If the gestational age of the fetal brain is in the early stage, the image of the fetal brain region is input into the early brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is mid-stage, the image of the fetal brain region is input into the mid-stage brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is late stage, the image of the fetal brain region is input into the late brain region segmentation model to obtain a segmented image of the fetal brain region.

[0009] Furthermore, the fetal brain region features include morphological features and radiomics features; wherein, radiomics features include first-order statistical features, texture features, and shape features.

[0010] Furthermore, the developmental offset is obtained in the following way: Based on the characteristics of each brain region of the fetus, a complete feature vector is constructed; The complete feature vector is input into the trained brain development equivalent regression model to obtain the fetal brain development equivalent gestational age. The developmental offset was obtained based on the equivalent gestational age of fetal brain development and the actual gestational age of the fetus.

[0011] Furthermore, the value of subtracting the actual gestational age of the fetus from the equivalent gestational age of fetal brain development is used as the developmental offset.

[0012] Furthermore, fetal brain development assessment results were obtained through the following methods: Based on the characteristics of each brain region of the fetus, key features of the fetus are obtained, and then key feature vectors are obtained. The key feature vector and developmental offset are input into the trained anomaly classifier to obtain the fetal brain development assessment result; wherein, the fetal brain development assessment result includes normal development or abnormal development, and if it is abnormal development, the assessment result also includes the abnormality probability.

[0013] Furthermore, the initial fetal MRI image is input into the trained brain region recognition model to obtain a confidence score; if the confidence norm is less than or equal to a set confidence threshold, the assessment of fetal brain development status is stopped.

[0014] Furthermore, the basic models of the segmented brain region segmentation models are all brain region segmentation models; the brain region segmentation models include encoders, decoders, and attentional jump connections; wherein, The encoder is used to compress and interpret the input image as hierarchical features; The decoder is used to recover the hierarchical features into a segmented image; The attention jump connection is used to connect the encoder and decoder for context-aware image reconstruction.

[0015] On the other hand, embodiments of the present invention provide a fetal brain development status assessment system, comprising: The image acquisition module is used to acquire MRI images of the fetus to be evaluated, and after preprocessing, obtain the initial fetal MRI image; A brain recognition module is used to obtain fetal brain region images based on the initial fetal MRI images and the trained brain region recognition model; The brain region segmentation and feature extraction module is used to obtain the gestational age stage of the fetal brain based on the fetal brain region image and the trained gestational age discriminator; and to obtain the fetal brain region segmentation image based on the fetal brain region image, the gestational age stage of the fetal brain and the trained segmented brain region segmentation model, and then extract the features of each fetal brain region. The fetal brain development assessment module is used to obtain the equivalent gestational age of fetal brain development based on the characteristics of each brain region of the fetus, and then to obtain the developmental offset based on the actual gestational age of the fetus, thereby obtaining the fetal brain development assessment result.

[0016] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: This invention provides a method and system for assessing fetal brain development. The method involves acquiring and preprocessing an initial fetal MRI image. Then, based on the initial fetal MRI image and a trained brain region recognition model, a fetal brain region image is obtained. Next, based on a trained gestational age discriminator, the gestational age stage of the fetal brain is determined. Subsequently, based on the fetal brain region image, the gestational age stage, and a trained segmented brain region segmentation model, a segmented fetal brain region image is obtained. Features of each fetal brain region are then extracted. Finally, based on these features, the equivalent gestational age of fetal brain development is determined. Finally, the developmental offset is calculated based on the actual gestational age, leading to the final assessment. The fetal brain development assessment results, through a one-stop analysis process, can quickly and comprehensively obtain various characteristic information of fetal brain development, greatly improving analysis efficiency and reducing the workload of doctors. Utilizing deep learning for brain region segmentation and feature analysis improves the accuracy and objectivity of the analysis and reduces the influence of human factors. By comprehensively considering morphological and radiomics characteristics, it can more comprehensively and deeply assess the status of fetal brain development. In particular, radiomics characteristics can capture changes in the microstructure of brain regions, providing a more reliable basis for clinical diagnosis and intervention. This helps to detect fetal brain developmental abnormalities early, take timely and appropriate treatment measures, reduce the incidence of neonatal neurological diseases, and improve the quality of the population.

[0017] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0018] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 This is a flowchart illustrating the fetal brain development status assessment method provided in Embodiment 1 of the present invention; Figure 2 A schematic diagram of a fetal brain region image provided in Embodiment 1 of the present invention, where the confidence norm is less than or equal to a set confidence threshold. Figure 3 A schematic diagram of a fetal brain region image provided in Embodiment 1 of the present invention, where the confidence norm is greater than a set confidence threshold. Figure 4 This is a schematic diagram of a fetal brain segmentation image provided in Embodiment 1 of the present invention. Detailed Implementation

[0019] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0020] Example 1 A specific embodiment of the present invention discloses a method for assessing fetal brain development status, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the MRI image of the fetus to be evaluated, and after preprocessing, obtain the initial fetal MRI image.

[0021] Specifically, preprocessing includes format conversion, resampling and size unification, intensity normalization, denoising, and enhancement. Among these, resampling and size unification involves resampling the image to isotropic resolution and adjusting it to a uniform size to meet the input requirements of the brain region recognition model; intensity normalization involves normalizing the pixel values ​​of the image to the range of [0,1] to eliminate signal differences caused by different scanners; and denoising and enhancement involve applying nonlocal mean or wavelet transform for denoising and using random rotation, flipping, brightness, or contrast adjustment.

[0022] S2. Based on the initial fetal MRI image and the trained brain region recognition model, obtain the fetal brain region image.

[0023] During the procedure, images of the fetal brain regions are obtained using the following methods: The initial fetal MRI image is input into a trained brain region recognition model to obtain a binary mask of the fetal brain region; wherein, the brain region recognition model is YOLOv8-seg; The initial fetal MRI image is masked based on the binary mask of the fetal brain region to obtain an image of the fetal brain region.

[0024] Specifically, the masking operation involves multiplying the initial fetal MRI image with a binary mask of the fetal brain region to obtain an image containing only the fetal brain region; where non-brain regions are converted to 0 or background values.

[0025] Preferably, the initial fetal MRI image is input into a trained brain region recognition model to obtain a confidence score; wherein, if the confidence score is less than or equal to a set confidence threshold, the assessment of fetal brain development status is stopped. Figure 2 The image shows the fetal brain region; if the confidence norm is greater than the set confidence threshold, the subsequent evaluation process continues, such as... Figure 3 The image shown depicts the fetal brain region.

[0026] In practice, the brain region recognition model is trained through the following process: Fetal MRI images were acquired after each diagnosis and preprocessed to construct the first dataset; the data in the first dataset consisted of the initial MRI images after each diagnosis. The brain boundaries of each initial MRI image after diagnosis in the first dataset were annotated and the data was processed to obtain the second dataset. The brain region recognition model was trained based on the second dataset to obtain a trained brain region recognition model.

[0027] Specifically, data processing includes geometric transformation, color transformation, and random occlusion; among which, geometric transformation includes random horizontal or vertical flipping, small-angle rotation (±10°), and scaling (0.9-1.1 times); color transformation includes random adjustment of brightness, contrast, and Gaussian noise to simulate different scanners and imaging conditions; random occlusion includes random rectangular occlusion, enabling the model to learn to be independent of local features.

[0028] Specifically, the second dataset is divided into training, validation, and test sets for training. The loss function during training is a composite loss function of bounding box regression loss, classification loss, and segmentation mask loss. The optimizer is AdamW. The initial learning rate is set to 1e-3, and a cosine annealing scheduling strategy is used to dynamically reduce the learning rate during training. The batch size is set to the largest possible value based on GPU memory (e.g., 16, 32). The number of training epochs is set to 300-500 epochs, and an early stopping function is enabled. When the validation set metric no longer improves within 20-30 consecutive epochs, training is automatically stopped to prevent overfitting.

[0029] It should be noted that the fetal MRI images after each diagnosis were acquired using T2-weighted images of the fetal brain from different models of MRI equipment, including different gestational weeks (12-41 weeks) and different sections (transverse, coronal, and sagittal). In addition, each fetal MRI image after diagnosis has a corresponding gestational week and fetal brain development status.

[0030] S3. Based on the fetal brain region image and the trained gestational age discriminator, the gestational age stage of the fetal brain is obtained.

[0031] Specifically, the gestational stages of the fetal brain include the early stage, the middle stage, and the late stage.

[0032] It should be noted that, in this embodiment, the entire gestational age range is divided into three consecutive stages based on the severity of brain morphological changes: the early stage, the middle stage, and the late stage. In the early stage, which occurs between 12 and 24 weeks of gestation, the fetal brain has smooth sulci and gyri, relatively large ventricles, and is in the early stages of tissue differentiation. In the middle stage, which occurs between 25 and 32 weeks of gestation, the fetal brain begins to form sulci and gyri rapidly, and the cortex thickens rapidly. In the late stage, which occurs between 33 and 41 weeks of gestation, the fetal brain's sulci and gyri become more complex, and the tissue structure resembles that of a newborn.

[0033] Specifically, the gestational age discriminator is a lightweight convolutional neural network classifier, such as ResNet or MobileNet; the input to the gestational age discriminator is an image of the fetal brain region, and the output is the gestational age stage of that fetal brain region image.

[0034] Specifically, the gestational age detector is trained in the following ways: The initial MRI images after diagnosis in the first dataset are input into the trained brain region recognition model to obtain images of each fetal brain region, thus constructing the third dataset. The images of each fetal brain region in the third dataset were labeled to obtain the fourth dataset; where the label is the gestational week stage. The gestational age discriminant was trained based on the fourth dataset, resulting in a well-trained gestational age discriminant.

[0035] S4. Based on the fetal brain region image, the gestational age of the fetal brain, and the trained segmented brain region segmentation model, a fetal brain region segmentation image is obtained, and then the features of each fetal brain region are extracted.

[0036] In implementation, the segmented brain region segmentation model includes an early brain region segmentation model, a mid-stage brain region segmentation model, and a late-stage brain region segmentation model; wherein, If the gestational age of the fetal brain is in the early stage, the image of the fetal brain region is input into the early brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is mid-stage, the image of the fetal brain region is input into the mid-stage brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is late stage, the image of the fetal brain region is input into the late brain region segmentation model to obtain a segmented image of the fetal brain region.

[0037] Understandably, the gestational age discriminator and the segmented brain region segmentation model ensure that the input image is always processed by the most suitable model, effectively solving the problem of performance degradation of a single model at both ends of the gestational age.

[0038] In practice, fetal brain region segmentation images are obtained through the following methods: The fetal brain region images are input into the corresponding trained stage brain region segmentation model to obtain the probability map of each fetal brain region; Based on the probability maps of different brain regions in the fetus, brain region segmentation and label maps are obtained; The brain region segmentation label is marked on the fetal brain region image to obtain the fetal brain region segmentation image.

[0039] Specifically, the probability maps of each brain region of the fetus are used to obtain brain region segmentation label maps based on ArgMax operations. That is, for each pixel position in the fetal brain region image, its probability value in the probability maps of each brain region of the fetus is compared, and the pixel position is assigned to the brain region where the probability map with the highest probability value is located.

[0040] Specifically, the fetal brain is divided into 21 brain regions, as shown in Table 1; the segmented images of the fetal brain regions are shown below. Figure 4 The image shown is a schematic diagram of two fetal brain region segmentation images.

[0041] Table 1. Brain regions divided in the fetal brain

[0042] In specific implementation, the basic model of the segmented brain region segmentation model is the brain region segmentation model; the brain region segmentation model includes an encoder, a decoder, and attentional jump connections; wherein... The encoder is used to compress and interpret the input image as hierarchical features; The decoder is used to restore the hierarchical features into probability maps of brain region segmentation; The attention jump connection is used to connect the encoder and decoder for context-aware image reconstruction.

[0043] Specifically, the encoder includes five downsampling stages, each consisting of two 3×3 convolutional layers, batch normalization, ReLU activation layers, and 2×2 max pooling layers, enabling the extraction of multi-scale features, from local details to global semantic information.

[0044] Specifically, the decoder includes five upsampling stages and an output layer. The five upsampling stages are symmetrical to the encoder. Each stage includes a 2×2 transposed convolution, two 3×3 convolutional layers, batch normalization, and a ReLU activation layer, which can gradually restore spatial details and generate fine segmentation results. Among them, each stage of the decoder is concatenated with the feature map of the corresponding encoder stage through attention jump connections.

[0045] Specifically, the attention-skipping connection performs 1×1 convolutions on both the encoder and decoder feature maps to unify the number of channels; after summing, it is activated by ReLU, then by a 1×1 convolution and Sigmoid activation to generate attention coefficients (between 0 and 1); the attention coefficients are multiplied by the original encoder feature map to obtain a weighted feature map; that is, in each upsampling stage of the decoder, its input consists of two parts: one is the upsampling result from the previous decoder layer, and the other is the feature map from the corresponding encoder layer, which is weighted by the attention gate mechanism. The two are concatenated and then fed into the subsequent convolutional layer; the attention-skipping connection enables the model to automatically focus on anatomical structural regions related to the segmentation task.

[0046] Preferably, auxiliary output branches are added at different levels of the decoder, including adding a 1×1 convolution after each upsampling stage to map the feature map to 21 categories and calculating a loss function for each auxiliary output, which can improve gradient flow, accelerate convergence, and improve model performance.

[0047] Specifically, the output layer is a 1×1 convolution with a Softmax activation function, which maps the 64-channel feature map to 21 channels, ensuring that the sum of the probabilities of the 21 channels for each pixel is 1, and outputs each probability map, with each location containing the classification of 21 brain regions.

[0048] Specifically, the brain region segmentation model for each stage is trained in the following way: Brain region segmentation and labeling were performed on the images of each fetal brain region in the third dataset to obtain brain region segmentation label maps, and the fourth dataset was constructed. Each data point in the fourth dataset consists of an image of each fetal brain region and its corresponding brain region segmentation label map. Based on the gestational age to which the images of each fetal brain region in the fourth dataset belong, the fourth dataset is divided into early dataset, mid-term dataset, and late dataset. After data augmentation of the early, intermediate, and late datasets, the brain region segmentation models were trained separately to obtain early, intermediate, and late brain region segmentation models.

[0049] Specifically, data augmentation includes: applying color dithering and Gaussian noise to early datasets to improve robustness to low contrast; and applying rotation and elastic deformation to mid- and late-stage datasets to enable the model to learn to understand the various complex shapes of brain sulci and gyri.

[0050] Specifically, the loss function used in the brain region segmentation model training process is a composite loss function of Dice Loss and Focal Loss; the optimizer is AdamW; the initial learning rate is set to 1e-3, and a cosine annealing scheduling strategy is adopted to dynamically reduce the learning rate during training; and for each brain region segmentation model, five-fold cross-validation is performed independently to ensure that the model performance is fully evaluated and stabilized within its respective gestational period.

[0051] Understandably, each segmentation model only needs to learn features within a relatively stable morphological range, making the task simpler and the model more focused, which can significantly improve the segmentation accuracy and robustness within each segment.

[0052] In practice, the fetal brain region features include morphological features and radiomics features; among which, radiomics features include first-order statistical features, texture features, and shape features.

[0053] Specifically, the morphological characteristics of fetal brain regions include absolute volume, proportional volume, symmetry ratio, cortical thickness, cortical surface area, and sulcus depth. Specifically, the first-order statistical features in radiomics include mean, variance, skewness, kurtosis, energy, and entropy; texture features include gray-level co-occurrence matrix, gray-level run matrix, and gray-level size region matrix; and shape features include sphericity, compactness, and surface area to volume ratio.

[0054] S5. Based on the characteristics of each brain region of the fetus, the equivalent gestational age of fetal brain development is obtained, and then the developmental offset is obtained according to the actual gestational age of the fetus, thereby obtaining the fetal brain development assessment results.

[0055] During implementation, the developmental offset is obtained in the following way: Based on the characteristics of each brain region of the fetus, a complete feature vector is constructed; The complete feature vector is input into the trained brain development equivalent regression model to obtain the fetal brain development equivalent gestational age. The developmental offset was obtained based on the equivalent gestational age of fetal brain development and the actual gestational age of the fetus.

[0056] Specifically, the developmental offset is calculated by subtracting the actual gestational age of the fetus from the equivalent gestational age of fetal brain development.

[0057] Specifically, brain development equivalent regression models use regression models that can handle high-dimensional features and are less prone to overfitting, such as gradient boosting regression trees (GBRT) or random forest regression models.

[0058] Specifically, the brain development equivalence regression model is trained in the following way: Images of fetal brain regions with normal fetal brain development and their corresponding brain region segmentation labels were selected from the fourth dataset. Based on the selected images of fetal brain regions and their corresponding brain region segmentation labels, brain region segmentation images of each fetus were obtained. Extract the features of each brain region from the segmented images of each fetal brain region to obtain the complete feature vector of each segmented image of the fetal brain region; The complete feature vectors of each fetal brain region segmentation image were labeled to construct the fifth dataset; where the gestational week label is the actual gestational week of the fetus. The brain development equivalent regression model is trained based on the fifth dataset to obtain a trained brain development equivalent regression model.

[0059] During implementation, in step S5, the fetal brain development assessment results are obtained through the following methods: Based on the characteristics of each brain region of the fetus, key features of the fetus are obtained, and then key feature vectors are obtained. The key feature vector and developmental offset are input into the trained anomaly classifier to obtain the fetal brain development assessment result; wherein, the fetal brain development assessment result includes normal development or abnormal development, and if it is abnormal development, the assessment result also includes the abnormality probability.

[0060] Specifically, the anomaly classifier uses support vector machine (SVM) or logistic regression.

[0061] Specifically, key fetal characteristics are screened using the following methods: Based on the images of each fetal brain region and their corresponding brain region segmentation labels in the fourth dataset, the segmentation images of each fetal brain region are obtained. Features of each brain region in the segmented images of each fetal brain region were extracted. The feature selection method with minimum redundancy and maximum relevance was adopted to select the feature subset most relevant to the abnormal state of fetal brain development as the key features of the fetus.

[0062] Specifically, based on the key fetal features extracted from the brain region segmentation images of each fetal brain region, the calculated developmental offset, and the corresponding diagnostic results, the abnormality classifier is trained to obtain a well-trained abnormality classifier.

[0063] Preferably, based on the assessment results, a detailed fetal brain development characteristic analysis report is automatically generated, including the actual gestational age of the fetus, the predicted equivalent gestational age for brain development, the developmental deviation, the brain region segmentation results, the values ​​of various characteristic parameters, the developmental status assessment conclusions, and suggestions, which are presented to clinicians in an intuitive and easy-to-understand manner.

[0064] Compared with existing technologies, this embodiment provides a method for assessing fetal brain development. It acquires MRI images of the fetus to be assessed, preprocesses them to obtain initial fetal MRI images, then uses these initial images and a trained brain region recognition model to obtain fetal brain region images. Next, using a trained gestational age discriminator, it determines the gestational age stage of the fetal brain. Then, based on the fetal brain region images, the gestational age stage, and a trained segmented brain region segmentation model, it obtains segmented fetal brain region images, extracts features from each brain region, and finally, based on these features, determines the equivalent gestational age of fetal brain development. Finally, it calculates the developmental offset based on the actual gestational age of the fetus. Obtaining fetal brain development assessment results through a one-stop analysis process allows for the rapid and comprehensive acquisition of various characteristic information about fetal brain development, greatly improving analysis efficiency and reducing the workload of doctors. Utilizing deep learning for brain region segmentation and feature analysis improves the accuracy and objectivity of the analysis, reducing the influence of human factors. By comprehensively considering morphological and radiomics characteristics, a more comprehensive and in-depth assessment of fetal brain development can be achieved. In particular, radiomics characteristics can capture changes in the microstructure of brain regions, providing a more reliable basis for clinical diagnosis and intervention. This helps in the early detection of fetal brain developmental abnormalities, timely implementation of corresponding treatment measures, reduction of the incidence of neonatal neurological diseases, and improvement of population quality.

[0065] Example 2 A specific embodiment of the present invention discloses a fetal brain development status assessment system, comprising: The image acquisition module is used to acquire MRI images of the fetus to be evaluated, and after preprocessing, obtain the initial fetal MRI image; A brain recognition module is used to obtain fetal brain region images based on the initial fetal MRI images and the trained brain region recognition model; The brain region segmentation and feature extraction module is used to obtain the gestational age stage of the fetal brain based on the fetal brain region image and the trained gestational age discriminator; and to obtain the fetal brain region segmentation image based on the fetal brain region image, the gestational age stage of the fetal brain and the trained segmented brain region segmentation model, and then extract the features of each fetal brain region. The fetal brain development assessment module is used to obtain the equivalent gestational age of fetal brain development based on the characteristics of each brain region of the fetus, and then to obtain the developmental offset based on the actual gestational age of the fetus, thereby obtaining the fetal brain development assessment result.

[0066] The specific implementation process of this invention can be found in the above method embodiments, and will not be repeated here.

[0067] Since this embodiment is based on the same principle as the above method embodiments, this system also has the corresponding technical effects of the above method embodiments.

[0068] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0069] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of assessing the state of fetal brain development, characterized by, Includes the following steps: Acquire fetal MRI images to be evaluated, perform preprocessing, and obtain initial fetal MRI images; Based on the initial fetal MRI image and the trained brain region recognition model, a fetal brain region image is obtained; the fetal brain region image is an image containing only the fetal brain region; Based on the fetal brain region images and the trained gestational age discriminator, the gestational age stage of the fetal brain is obtained; Based on the fetal brain region image, the gestational age stage of the fetal brain, and a trained segmented brain region segmentation model, a segmented fetal brain image is obtained, and then features of each fetal brain region are extracted. The gestational age stage of the fetal brain includes early, mid, and late stages; the segmented brain region segmentation model includes early, mid, and late brain region segmentation models; the basic model of each segmented brain region segmentation model is a brain region segmentation model; the brain region segmentation model includes an encoder, a decoder, and attentional jump connections; the encoder is used to compress and interpret the input image into hierarchical features; the decoder is used to recover the segmented image from the hierarchical features; the attentional jump connections are used to connect the encoder and decoder for context-aware image reconstruction; the segmented fetal brain image is obtained by segmenting and labeling each brain region on the fetal brain region image. Based on the characteristics of each brain region of the fetus, the equivalent gestational age of fetal brain development is obtained, and then the developmental deviation is obtained according to the actual gestational age of the fetus, thus obtaining the fetal brain development assessment results.

2. The method of fetal brain development status assessment according to claim 1, characterized in that, Images of fetal brain regions are obtained through the following methods: The initial fetal MRI image is input into a trained brain region recognition model to obtain a binary mask of the fetal brain region; wherein, the brain region recognition model is YOLOv8-seg; The initial fetal MRI image is masked based on the binary mask of the fetal brain region to obtain an image of the fetal brain region.

3. The method for assessing fetal brain development status according to claim 1, characterized in that, If the gestational age of the fetal brain is in the early stage, the image of the fetal brain region is input into the early brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is mid-stage, the image of the fetal brain region is input into the mid-stage brain region segmentation model to obtain a segmented image of the fetal brain region. If the gestational age of the fetal brain is late stage, the image of the fetal brain region is input into the late brain region segmentation model to obtain a segmented image of the fetal brain region.

4. The method of fetal brain development assessment according to claim 1, wherein, The fetal brain region features include morphological features and radiomics features; among which, radiomics features include first-order statistical features, texture features, and shape features.

5. The method for assessing fetal brain development status according to claim 4, characterized in that, The developmental offset is obtained in the following way: Based on the characteristics of each brain region of the fetus, a complete feature vector is constructed; The complete feature vector is input into the trained brain development equivalent regression model to obtain the fetal brain development equivalent gestational age. The developmental offset was obtained based on the equivalent gestational age of fetal brain development and the actual gestational age of the fetus.

6. The method for assessing fetal brain development status according to claim 5, characterized in that, The developmental offset is calculated by subtracting the actual gestational age of the fetus from the equivalent gestational age of fetal brain development.

7. The method for assessing fetal brain development status according to claim 5, characterized in that, Fetal brain development assessment results are obtained through the following methods: Based on the characteristics of each brain region of the fetus, key features of the fetus are obtained, and then key feature vectors are obtained. The key feature vector and developmental offset are input into the trained anomaly classifier to obtain the fetal brain development assessment result; wherein, the fetal brain development assessment result includes normal development or abnormal development, and if it is abnormal development, the assessment result also includes the abnormality probability.

8. The method for assessing fetal brain development status according to claim 2, characterized in that, The initial fetal MRI image is input into the trained brain region recognition model to obtain a confidence score; if the confidence norm is less than or equal to the set confidence threshold, the assessment of fetal brain development status is stopped.

9. A fetal brain development status assessment system, characterized in that, include: The image acquisition module is used to acquire MRI images of the fetus to be evaluated, and after preprocessing, obtain the initial fetal MRI image; The brain recognition module is used to obtain a fetal brain region image based on the initial fetal MRI image and the trained brain region recognition model; the fetal brain region image is an image containing only the fetal brain region; The brain region segmentation and feature extraction module is used to obtain the gestational age stage of the fetal brain based on the fetal brain region image and a trained gestational age discriminator; it also obtains a segmented fetal brain image based on the fetal brain region image, the gestational age stage of the fetal brain, and a trained segmented brain region segmentation model, and then extracts features of each fetal brain region; wherein, the gestational age stage of the fetal brain includes an early stage, a mid-stage, and a late stage; the segmented brain region segmentation model includes an early brain region segmentation model, a mid-stage brain region segmentation model, and a late-stage brain region segmentation model; the basic model of the segmented brain region segmentation model is a brain region segmentation model; the brain region segmentation model includes an encoder, a decoder, and attention jump connections; wherein, the encoder is used to compress and interpret the input image into hierarchical features; the decoder is used to restore the hierarchical features into a segmented image; the attention jump connections are used to connect the encoder and decoder to perform context-aware image reconstruction; the segmented fetal brain image is an image obtained after segmenting and labeling each brain region on the fetal brain region image; The fetal brain development assessment module is used to obtain the equivalent gestational age of fetal brain development based on the characteristics of each brain region of the fetus, and then to obtain the developmental offset based on the actual gestational age of the fetus, thereby obtaining the fetal brain development assessment result.

Images