A method and system for predicting the development of early cortical folds in the human brain based on a neural network model

Abstract

The present invention discloses a method and system for predicting the development of early cortical folds in the human brain based on a neural network model, which falls into the interdisciplinary field of artificial intelligence and brain science. The method extracts a three-dimensional skeleton of cortical folds using brain negative model skeleton technology, and uses a β-VAE model to capture the complex pattern of folds changing with gestational age, achieving accurate predictions from 30 to 40 weeks of gestation. The method uses a large amount of MRI imaging data to train the model to ensure the accuracy of the prediction results. The system modular design of the method improves scalability and maintainability, and the computer implementation ensures real-time and accuracy. The present invention solves the problem of the lack of neurogenesis heterogeneity information in existing models, provides a new approach for studying the development of early cortical folds in the human brain, and has broad application prospects, such as evaluating the neurodevelopment of premature infants and early diagnosis of diseases.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary field of artificial intelligence and brain science, and specifically relates to a method and system for predicting the development of early cortical folds in the human brain based on a neural network model. Background Art

[0002] The formation of gyri (outward-facing convex areas) and sulci (inward-facing concave areas) on the complex cortical surface is a complex and delicate developmental process that unfolds gradually between 16 and 40 weeks of gestation. The location of primary gyri and sulci, which appear between 16 and 29 weeks, is relatively consistent across the population. Secondary and tertiary gyri and sulci appear in the middle and late stages of cortical development (after 29 weeks). These structures vary significantly between individuals, influenced by a variety of factors, including genetics and the environment. This diversity complicates the study of the developmental mechanisms of these structures.

[0003] The origins of some neurodevelopmental disorders associated with abnormal folding patterns are not yet fully understood. Recent studies have shown that the formation of cortical folds is regulated by multiple molecular and cellular processes, including neuronal proliferation, migration, differentiation, and synapse formation. In addition, mechanical force also plays an important role. The difference in growth rate of the cerebral cortex leads to an imbalance of tension between the inner and outer layers, which in turn triggers the formation of folds.

[0004] In recent years, teams both domestic and international have developed a variety of mechanical models based on diverse mechanical assumptions to simulate the formation of cortical folds. However, these models lack information on the spatial heterogeneity of neurogenesis, such as regional differential growth rates. Consequently, they are unable to fully simulate the complex changes in the cerebral cortex during development. Simulation results from these models consistently indicate that they do not accurately reflect the common topological structure of the human brain. Dissection data indicate that between 30 and 38 weeks of gestation, the lateral parietal, temporal, and occipital regions experience significantly higher rates of cortical expansion than the medial and insular regions. This regional difference in expansion rate is a key factor contributing to the diversity of cortical folding patterns. Furthermore, during the period of rapid cortical expansion (29-38 weeks of gestational age), the growth patterns of different regions vary significantly over time. For example, the expansion rate of the primary sensory, motor, and visual cortices, as well as the insula, slows over time, while the expansion rate of the lateral temporal lobe accelerates.

[0005] The development of fetal cortical folding is a complex and delicate process. In the field of neuroscience, understanding fetal brain development, especially the dynamic changes in cortical folding, is crucial for diagnosing early neurodevelopmental abnormalities. The dynamic changes in cortical folding involve regulatory factors at multiple levels, necessitating the comprehensive application of multiple methods and techniques to more fully reveal its developmental mechanisms and promote progress in related fields. However, because extremely premature infants (e.g., around 30 weeks) have minimal cortical folding at birth, directly observing their development presents challenges. Summary of the Invention

[0006] In order to address the technical problem that existing physical models lack spatial heterogeneity information on neurogenesis phenomena when simulating the development of fetal cerebral cortical folds, and such information is difficult to obtain directly in the human brain, resulting in the simulation results being unable to accurately reflect the actual topological structure of the human brain, the present invention aims to provide a method and system for learning and predicting the development of early human brain cortical folds based on a neural network model.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The present invention provides a method for predicting the development of early cortical folds in the human brain based on a neural network model, comprising:

[0009] S1, obtain imaging data of the adult brain and the brain of premature infants at 30 and 40 weeks of gestational age;

[0010] S2, based on the brain negative skeleton technology, extracts the three-dimensional skeleton structure of cortical folds from the image data obtained from S1;

[0011] S3, input the wrinkle skeleton data of S2 at 30 weeks of gestation into the pre-trained β variational autoencoder model to predict the wrinkle skeleton morphology at 40 weeks of gestation;

[0012] S4 outputs the predicted cortical folds at 40 weeks of gestation, which are highly similar to the actual brain fold morphology.

[0013] The brain imaging data are MRI T1w imaging data of extremely premature infants:

[0014] The three-dimensional skeleton structure of cortical folds extracted by S2 includes:

[0015] S21, preprocess the image data in S1 using the brain extraction algorithm to extract the cerebral cortex area;

[0016] S22, based on the cerebral cortex area image data extracted in S21, reconstructing a three-dimensional surface model of the cerebral cortex using three-dimensional reconstruction technology;

[0017] S23, converting the three-dimensional surface model of the cerebral cortex reconstructed in S22 into a negative model;

[0018] S24, applying the skeleton extraction algorithm on the negative model of S23 to extract the three-dimensional skeleton structure of the cortical folds;

[0019] S25, post-processing the three-dimensional skeleton structure extracted by S24, optimizing and reconstructing it, and obtaining the three-dimensional skeleton structure of the cortical folds.

[0020] The S3 pre-trained β variational autoencoder model includes:

[0021] S31, input the three-dimensional skeleton structure of the cortical folds extracted by S2 into the β variational autoencoder model;

[0022] S32, build the β-VAE model framework, which includes an encoder and a decoder. The encoder is responsible for mapping the input 3D cortical geometry information to the latent space, while the decoder is responsible for reconstructing the 3D cortical structure from the latent space.

[0023] S33, using preprocessed adult brain MRI data as a training set, the β-VAE model was trained. The model parameters were adjusted by minimizing the reconstruction error and the KL divergence between the latent variable distribution and the prior distribution. The model was trained to learn the skeleton topology of adult brain folds and achieve adult brain fold data reconstruction. The reconstruction error was calculated using the cross entropy loss, and the prior distribution was a central Gaussian distribution. The pretrained model was evaluated using the validation set, and the pretrained machine learning model was obtained based on the model evaluation indicators.

[0024] S34, using the pre-processed premature infant image data as a training set, retrained the β-VAE model pre-trained in S33. By inputting the cortical fold skeleton data of infants at 30 weeks of gestation, the model learned and predicted the fold morphology of these areas at 40 weeks of gestation;

[0025] S35, data partitioning and model optimization, during the training process, the data set is divided into training set, validation set and test set to adjust and optimize model parameters, evaluate the performance of the model, and finally obtain the optimal machine learning model; determine the best model hyperparameter combination, the hyperparameter combination includes β value, latent space size, batch size, number of training rounds and learning rate.

[0026] Furthermore, the S35 hyperparameter combination is: β value range is 1~8, latent space size range is 2~500, batch size range is 4~16, number of training rounds range is 500~2000, and learning rate range is 1e-4~1e-2.

[0027] Furthermore, the optimal combination of hyperparameters is: β value is 2, latent space size is 300, batch size is 8, number of training rounds is 1000, and learning rate is 1e-3.

[0028] The hyperparameter combination of the optimization model adopts any one of grid search, random search or Bayesian optimization.

[0029] The system for implementing the above-mentioned method for predicting the development of early cortical folds in the human brain based on a neural network model comprises:

[0030] Data acquisition and processing module, used to acquire imaging data of 30-week fetuses and adult brains, and pre-process the imaging data;

[0031] 3D skeleton extraction module, used to extract the 3D skeleton structure of cortical folds;

[0032] The model training and prediction module is used to input the wrinkle skeleton data at 30 weeks of gestation into the trained model to predict the wrinkle skeleton morphology at 40 weeks of gestation;

[0033] Hyperparameter optimization module, used to obtain the best hyperparameter combination;

[0034] The result evaluation and validation module is used to verify the accuracy of the model in predicting the cortical fold morphology at 40 weeks of gestation and its effectiveness in practical applications;

[0035] The user interaction and output module is used to output the prediction results to the user in graphical or tabular form.

[0036] The present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting the development of early cortical folds in the human brain based on a neural network model are implemented.

[0037] The present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the steps of the method for predicting the development of early cortical folds in the human brain based on a neural network model.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention provides a method for predicting the development of early cortical folds in the human brain based on a neural network model. This method uses brain negative skeleton technology to effectively extract the three-dimensional skeleton structure of cortical folds from complex imaging data. This method not only simplifies the complexity of data processing but also retains key structural information, providing a solid foundation for subsequent model training and prediction. Using a β-variational autoencoder (β-VAE) model, deep learning is used to capture the complex geometric characteristics of cerebral cortical folds and their patterns of change with gestational age. This model can accurately predict the fold morphology at 40 weeks of gestation from fold skeleton data at 30 weeks of gestation. The predicted results are highly similar to the actual brain fold morphology, improving the accuracy of the prediction. The method of the present invention has broad application prospects in fields such as neuroscience, medical imaging, and clinical medicine. It can not only be used to assess the neurodevelopmental status of premature infants, but also provide important reference information for the early diagnosis and intervention of related diseases. In addition, the method can be extended to other research fields related to brain development, such as cognitive science and psychology.

[0040] Furthermore, the method of the present invention relies on a large amount of MRI imaging data, including MRI data of 30-week gestational age and adult brains, as well as MRI data of premature infants. This data provides a rich training sample for the model, enabling it to learn the general patterns and individual differences in the changes of cerebral cortical folds with gestational age. During the model training process, strategies such as data partitioning, model evaluation, and hyperparameter optimization are adopted. By adjusting hyperparameters such as β value, latent space dimension, and batch size, the performance of the model can be further optimized to make it more suitable for specific prediction tasks. This flexibility enables the model to maintain high prediction accuracy in different data sets and application scenarios.

[0041] The system provided by the present invention includes multiple modules such as data acquisition and processing, three-dimensional skeleton extraction, model training and prediction, hyperparameter optimization, result evaluation and verification, and user interaction and output. These modules are independent of each other but work closely together to achieve the function of predicting the development of early cortical folds in the human brain based on a neural network model. This design improves the scalability and maintainability of the system. By implementing the method of the present invention as a computer program and storing it on a computer-readable storage medium, it can be quickly deployed and run on a computer device. The powerful computing power of the computer device and the efficient algorithm implementation ensure the real-time and accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a method for predicting the development of early cortical folds in the human brain based on a neural network model according to the present invention;

[0043] Figure 2 Schematic diagram of the method for identifying human cerebral cortical folds based on TI-weighted MRI images according to the present invention;

[0044] Figure 3 The cortical fold morphology of part of the 40-week fetus output by the model trained by the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0046] The present invention is described in further detail below with reference to the accompanying drawings:

[0047] like Figure 1As shown, a process and method diagram for generating a three-dimensional brain fold skeleton and training a β variational autoencoder model is provided, including:

[0048] 1. Data Acquisition and Preprocessing

[0049] (1) Data acquisition

[0050] Data Sources: The model was pre-trained using T1-weighted MRI brain images from the Human Connectome Project (HCP) dataset of 900 healthy adults, aged 22 to 36 years, acquired on a Siemens Skyra Connectom scanner at an isotropic resolution of 0.7 mm. The model was re-trained using T1-weighted MRI brain images from 71 premature infants from the Utrecht dataset, acquired at 30 and 40 weeks of gestational age.

[0051] Data partitioning: 80% of the data is randomly selected as the training set for training the β-Variational Autoencoder (β-VAE) model; 10% of the data is used as the validation set to adjust model parameters and prevent overfitting; the remaining 10% is used as the test set to evaluate the generalization ability and prediction accuracy of the model.

[0052] (2) Data preprocessing

[0053] See attached Figure 2 Based on the negative skeleton technology of cerebral cortical folds, the three-dimensional skeleton structure of cortical folds was extracted, including:

[0054] 3D calibration: The DISCO+DARTEL algorithm is used to perform nonlinear calibration on the original T1-wMRI images to eliminate geometric differences and deformations between different scans and ensure the accuracy of subsequent analysis;

[0055] Brain region identification and segmentation: Through brain recognition algorithms, different brain regions are automatically identified and segmented, clearly displaying the different brain regions and providing a basis for subsequent 3D reconstruction;

[0056] Gray matter and white matter segmentation: Based on brain region segmentation, gray matter and white matter layers are further distinguished to more accurately analyze cortical folds. The segmentation images of the cerebral hemispheres and gray and white matter clearly define the boundaries between gray and white matter.

[0057] Negative model recognition and skeleton extraction: An automatic brain fold skeleton recognition algorithm is applied to the segmented white matter layer to identify the negative model structure of the folds and extract the 3D skeleton. This step is the core of extracting the 3D structure of cortical folds. The cerebral cortical images of the left and right hemisphere posterior shells are the results of rendering or visualization after skeleton extraction.

[0058] Post-processing and optimization: The extracted 3D skeleton structure is post-processed, including smoothing, denoising, and optimized reconstruction, to obtain a more accurate and clear 3D skeleton structure of cortical folds. The automatically labeled fold map is used for further analysis or visualization of specific areas.

[0059] 2. Model Pre-training

[0060] Pre-training phase: First, a neural network β variational autoencoder model is pre-trained using a "negative 3D skeleton" of the left and right hemispheres extracted from adult brain MRI. The goal of this phase is to enable the model to learn the geometric shape and spatial distribution of the cerebral cortex, achieve accurate reconstruction of the fold skeleton, and provide a basis for subsequent research on the development process of early cortical folds in the human brain.

[0061] Model Structure: A β-Variational Autoencoder (β-VAE) model is used. Compared to traditional variational autoencoders, β-VAE introduces a regulation term β to change the weight distribution of the loss function in the model structure, placing greater emphasis on learning more interpretable latent variables. This model performs particularly well in unsupervised learning and image generation tasks. By adjusting the value of β, a trade-off can be made between reconstruction accuracy and the interpretability of the latent representation. When β > 1, the model places greater emphasis on decoupling. While this may sacrifice some reconstruction quality, it has a significant advantage in generating more interpretable representations.

[0062] 3. Model Retraining

[0063] Target areas: The central sulcus of the fetal brain and its adjacent precentral gyrus and postcentral gyrus are the focus of research. These areas play an important role in brain function, and changes in their fold morphology can reflect the early development of the brain.

[0064] Input and Output: The 3D structure of cortical folds is represented using a "negative 3D skeleton of the brain." The cortical fold skeleton data at 30 weeks of gestation is used as input, and the model is trained to predict the fold skeleton morphology of the corresponding region at 40 weeks of gestation.

[0065] 4. Hyperparameter Optimization

[0066] Optimization process: Using the Adam (Adaptive Moment Estimation) optimizer, a gradient descent optimization algorithm, and an adaptive learning rate adjustment mechanism, the optimal hyperparameter combination was determined through tuning the initial values ​​of the model hyperparameters and extensive experimentation. This included the β value, the size of the latent space, the batch size, the number of training rounds, and the learning rate. In this paper, the final β value was 2 (tested between 1 and 8), which performed well in balancing reconstruction accuracy and interpretability of the latent representation. The dimensionality of the latent space was 300 (tested between 2 and 500), which fully captures the complex variations in brain fold morphology. The batch size was 8 (tested between 4 and 16) to ensure that the model was neither too conservative nor too aggressive during training. The number of training rounds was 1000 (tested between 500 and 2000) to ensure sufficient model convergence and avoid overfitting. The learning rate was 1e-3 (tested between 1e-4 and 1e-2), which ensured stable and efficient parameter updates during training.

[0067] Regularization function: By utilizing the regularization function of the β-VAE model, by balancing the reconstruction error and the decoupling of the latent space, the overfitting problem caused by the small number of training samples is alleviated, and the generalization ability of the model is improved.

[0068] 5. Model evaluation and results analysis

[0069] Model output: The accuracy of the model was evaluated by comparing the model-output cortical fold morphology at 40 weeks of gestation with the actual (target) brain fold morphology. Figure 3 As shown, some of the prediction results output by the model are highly similar to the actual results, which proves the accuracy and reliability of the model in predicting the fold morphology of the cerebral cortex of premature infants.

[0070] Input: The input refers to the skeleton data of the central sulcus and the anterior-posterior central sulcus of premature infants at 30 weeks of gestational age. These data are used as input to the model to train the model to predict future fold morphology.

[0071] Output: The output is the skeletal morphology of the central sulcus and the anterior-posterior central sulcus at 40 weeks of gestation, predicted by the model based on the input data. These predictions are displayed in the form of 3D skeleton diagrams in the image, and each skeleton diagram represents the prediction result for a specific individual.

[0072] Target: The target refers to the actual, known cortical fold morphology of the corresponding region at 40 weeks of gestational age, which is used to compare with the model's prediction results to evaluate the model's accuracy.

[0073] By the attached Figure 3From the data, we can see that the skeleton diagrams of individual 1 to individual 6 each represent the cortical fold morphology of different premature infants under the model prediction. The differences in the position and geometry of these skeleton diagrams reflect the individual differences in brain development of different infants. By comparing the model output (predicted results) with the target (actual results), the accuracy and reliability of the model in predicting the fold morphology of the cerebral cortex of premature infants can be evaluated. The model outputs some examples of cortical fold morphology at 40 weeks of gestation. The model is able to capture the individual differences in brain development of different premature infants and accurately predict the changing trend of fold morphology in the future. This result not only verifies the effectiveness of the model, but also provides a new perspective and tool for understanding the process of brain development. For example, the predicted results of individual 1 and individual 4 are compared with the actual (target) results. Comparison of brain fold morphology shows that the model's prediction accuracy is high, indicating that the model has high accuracy in predicting the fold morphology of the cerebral cortex of premature infants and can capture key features of brain development. Because the model adopts the β-VAE structure and improves performance through hyperparameter optimization, it is able to learn effective latent representations with limited training samples and avoid overfitting problems. This demonstrates the effectiveness and robustness of the model in complex biological data processing tasks. The model can still learn effective latent representations and avoid overfitting problems with limited training samples, thanks to the regularization function of the β-VAE model and the fine-tuning of hyperparameter optimization. These results show that the model performs well in complex biological data processing tasks and has broad application prospects and potential.

[0074] In summary, this study successfully constructed a β-VAE model capable of predicting cortical fold morphology in premature infants through a series of steps, including data acquisition, preprocessing, model pretraining, retraining, hyperparameter optimization, model evaluation, and results analysis. This model not only provides new insights and tools for understanding brain development but also offers strong support for clinical diagnosis and treatment in related fields.

[0075] The present invention is described with reference to flowcharts of methods, algorithms, and experiments according to embodiments of the invention.

[0076] It should be noted that, in this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0077] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0078] The above content is only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for predicting the development of early cortical folds in the human brain based on a neural network model, characterized in that: include: S1, obtain imaging data of the adult brain and the brain of premature infants at 30 and 40 weeks of gestational age; S2, based on the brain negative skeleton technology, extracts the three-dimensional skeleton structure of cortical folds from the image data obtained from S1; The three-dimensional skeleton structure of the cortical folds is extracted including: S21, preprocess the image data in S1 using the brain extraction algorithm to extract the cerebral cortex area; S22, based on the cerebral cortex area image data extracted in S21, reconstructing a three-dimensional surface model of the cerebral cortex using three-dimensional reconstruction technology; S23, converting the three-dimensional surface model of the cerebral cortex reconstructed in S22 into a negative model; S24, applying the skeleton extraction algorithm on the negative model of S23 to extract the three-dimensional skeleton structure of the cortical folds; S25, post-processing the three-dimensional skeleton structure extracted by S24, optimizing and reconstructing it, and obtaining the three-dimensional skeleton structure of the cortical folds; S3, using the adult wrinkle skeleton data of S2 as input to pre-train a β variational autoencoder model to achieve wrinkle reconstruction, and then using the wrinkle skeleton data at 30 weeks of gestation as input to the pre-trained β variational autoencoder model to predict the wrinkle skeleton morphology at 40 weeks of gestation; S4 outputs the predicted cortical folds at 40 weeks of gestation, which are highly similar to the actual brain fold morphology.

2. The method for predicting the development of early cortical folds in the human brain based on a neural network model according to claim 1, characterized in that: The S1 brain image data are T1-weighted MRI image data of adults and extremely premature infants.

3. The method for predicting the development of early cortical folds in the human brain based on a neural network model according to claim 1, characterized in that: The S3 β variational autoencoder model includes: S31, input the three-dimensional skeleton structure of the cortical folds extracted in S2; S32, build the β-VAE model framework, which includes an encoder and a decoder. The encoder is responsible for mapping the input 3D cortical geometry information to the latent space, while the decoder is responsible for reconstructing the 3D cortical structure from the latent space. S33, using preprocessed adult brain MRI data as a training set, the β-VAE model was trained. The model parameters were adjusted by minimizing the reconstruction error and the KL divergence between the latent variable distribution and the prior distribution. The model was trained to learn the skeleton topology of adult brain folds and achieve adult brain fold data reconstruction. The reconstruction error was calculated using the cross entropy loss, and the prior distribution was a central Gaussian distribution. The pretrained model was evaluated using the validation set, and the pretrained machine learning model was obtained based on the model evaluation indicators. S34, using the pre-processed premature infant image data as a training set, retrained the β-VAE model pre-trained in S33. By inputting the cortical fold skeleton data of infants at 30 weeks of gestation, the model learned and predicted the fold morphology of these areas at 40 weeks of gestation; S35, data partitioning and model optimization, during the training process, the data set is divided into training set, validation set and test set, the model parameters are adjusted and optimized, the performance of the model is evaluated, and finally the optimal machine learning model is obtained; the optimal model hyperparameter combination is determined, and the hyperparameter combination includes β value, latent space size, batch size, number of training rounds and learning rate.

4. The method for predicting the development of early cortical folds in the human brain based on a neural network model according to claim 3, characterized in that: The S35 hyperparameter combinations are: β value range is 1~8, latent space size range is 2~500, batch size range is 4~16, number of training rounds range is 500~2000, and learning rate range is 1e-4~1e-2.

5. The method for predicting the development of early cortical folds in the human brain based on a neural network model according to claim 4 is characterized in that: The optimal combination of hyperparameters is: β value of 2, latent space size of 300, batch size of 8, number of training rounds of 1000, and learning rate of 1e-3.

6. The method for predicting the development of early cortical folds in the human brain based on a neural network model according to claim 3, characterized in that: The hyperparameter combination of the optimization model adopts any one of grid search, random search or Bayesian optimization.

7. A system for implementing the method for predicting the development of early cortical folds in the human brain based on a neural network model as described in any one of claims 1 to 6, characterized in that: include: The data acquisition and processing module is used to acquire imaging data of 30-week fetal and adult brains, and perform nonlinear calibration and reconstruction preprocessing based on brain sulci; 3D skeleton extraction module, used to extract the 3D skeleton structure of cortical folds; The model training and prediction module is used to input the wrinkle skeleton data at 30 weeks of gestation into the pre-trained model to predict the wrinkle skeleton morphology at 40 weeks of gestation; Hyperparameter optimization module, used to obtain the best hyperparameter combination; The result evaluation and validation module is used to verify the accuracy of the model in predicting the cortical fold morphology at 40 weeks of gestation and its effectiveness in practical applications; The user interaction and output module is used to output the prediction results to the user in graphical or tabular form.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting the development of early cortical folds in the human brain based on a neural network model as described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the development of early cortical folds in the human brain based on a neural network model as described in any one of claims 1 to 6 are implemented.

Images