Method for predicting cerebral palsy risk of BIPI child patient

Abstract

The invention provides a method for predicting the cerebral palsy risk of a BIPI child patient. The method comprises the steps of obtaining a data set; the data set is a brain MRI image of the BIPI child patient; building an automatic focus segmentation model of the BIPI child patient; obtaining a focus segmentation result; constructing a KD grading prediction model; a Kidokoro scoring result is obtained; constructing a focus feature extraction network model; automatically extracting image features of the brain MRI image of the BIPI child patient corresponding to the data set; constructing a brain paralysis risk prediction model of the BIPI child patient, and taking the GMs evaluation result after quantitative conversion as the output of the prediction model; integrating the image features of the brain MRI image corresponding to the data set, the quantized and converted Kidokoro scoring result and the perinatal period clinical index data into a combination, and taking the combination as the input of the prediction model; and obtaining a cerebral palsy risk prediction result of the BIPI child patient. According to the BIPI child patient cerebral palsy risk prediction model, multi-dimensional factors are incorporated to comprehensively evaluate the BIPI child patient cerebral palsy risk, and the accuracy and practicability of early prediction are innovatively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cerebral palsy risk prediction. Specifically, it relates to a method for predicting the cerebral palsy risk of BIPI children. Background Art

[0002] Brain Injury In Premature Infants (BIPI) refers to various degrees of cerebral ischemia or / and hemorrhagic and inflammatory damage in premature infants caused by various perinatal and intrapartum pathological factors, and corresponding symptoms and signs appear clinically. In severe cases, it can lead to neurological sequelae or even death. With the increasing total survival rate of premature infants, the incidence of cerebral palsy in BIPI children has been rising year by year. Research statistics in China show that cerebral palsy caused by premature complications accounts for 29.13‰ of newborns. Prevention and early intervention are the keys to reducing the risk of cerebral palsy, but the early diagnosis of cerebral palsy is often very difficult.

[0003] The level of brain injury in BIPI children is closely related to the prognosis of cerebral palsy risk. The level of brain injury in BIPI children is positively correlated with the clinical prognosis. However, the determination of the level of brain injury in BIPI children lacks corresponding clinical signs and indicators, which has become a major clinical problem. Studies by Kang O, M N B, Kekky L, etc. have confirmed that the MRI scoring system can quantitatively determine the level of brain injury and evaluate the maturity of brain development, and has good predictive value for the long-term prognosis of BIPI children, including cognitive, motor, and language outcomes. The traditional MRI scoring system only defines the scoring rules for grading the degree of brain injury in full-term infants with hypoxic-ischemic encephalopathy (HIE for short). The two most commonly used scoring systems are the Barkovich score and the Neonatal Research Network of the National Institute of Child Health and Human Development (NICHD NRN) score. These two scoring systems have great limitations and only analyze the injuries of the cerebral white matter (WM) and deep gray matter structures, and do not include the injuries of other regions in the scoring rules. The Kidokoro (KD for short) scoring system not only covers a comprehensive range of injury areas, but also conducts a detailed score assessment on the injury area patterns of each region. It also incorporates cortical fold development, myelin maturation, and volume changes of important organizational structures after injury into the scoring system. The most crucial point is that the validation group of this KD scoring system is applicable to premature infants. Therefore, this KD scoring system is more comprehensive, objective, and has a wider applicable population in clinical applications, and the accuracy of determining the degree of injury by this KD scoring system has been clinically verified and recognized.

[0004] Although the KD scoring system has high clinical application value, the high complexity of the KD scoring system and the high time-consuming nature of manual scoring pose great challenges to the scorers. The KD scoring system requires senior pediatric neuroimaging experts who are professional and familiar with the scoring criteria to complete it. Moreover, the brain region structure measurement work in the KD scoring system is time-consuming. It takes 10 - 15 minutes to complete the complete scoring of one child. Facing the huge clinical workload, the clinical popularization of the KD scoring system is difficult to achieve and is only limited to children with special conditions and urgent needs. The emergence of artificial intelligence makes it possible to break through this clinical bottleneck. Currently, although there are more and more studies on premature infants, there has been no report on constructing an intelligent grading AI model for the degree of injury based on a scoring system.

[0005] In addition, neurodevelopmental assessment follow-up is the main clinical method relied on for the clinical diagnosis of cerebral palsy and the evaluation of the curative effect after the intervention treatment of neurodevelopmental abnormalities. Traditional clinical assessment tools include multiple assessment methods such as the Brazelton Neonatal Behavioral Assessment Scale, the Amiel-Tison Neonatal Neurological Test, the Bayley Scales of Infant Development, and the Neonatal Behavioral Neurological Assessment of 20 Items (abbreviated as NBNA). However, these assessment tools all have limitations such as long time-consuming, inconsistent accuracy, and late diagnosis time. Especially in terms of the diagnosis time, it is often not possible to diagnose until the child is 2 years old, which affects the best timing of early rehabilitation treatment. General movements (GMs) assessment, as an internationally recognized new type of movement assessment method, is highly regarded clinically because of its accuracy, simplicity, non-invasiveness, and high predictability. However, although the GMs assessment method has high accuracy, its result acquisition only depends on the clinical signs that the child has already shown, and it requires professional GMs assessors to conduct the assessment. And through clinical GMs assessment, the high-risk indication time for the child to have cerebral palsy and movement disorders is at the earliest 5 months after birth, and for many children, the time will be extended to 1 year. Although the GMs prediction time for cerebral palsy is earlier than that of other assessment tools, the best time for early intervention of premature infant brain injury is within 1 month after birth. At this time, the probability of brain plasticity in children is high and the clinical intervention result is the best.

[0006] Therefore, in fact, in view of a series of defects existing in the related technologies for predicting the risk of cerebral palsy in BIPI children, what we clinically urgently need is a method for early predicting the risk of cerebral palsy in BIPI children. Summary of the Invention

[0007] The present invention aims to at least solve one of the technical problems existing in the prior art or related technologies.

[0008] For this reason, the purpose of the present invention is to propose a method for predicting the risk of cerebral palsy in BIPI children.

[0009] To achieve the above object, the technical solution of the present invention provides a method for predicting the risk of cerebral palsy in BIPI children. The method for predicting the risk of cerebral palsy in BIPI children includes: Step S1: Obtain a data set; wherein, the data set is the brain MRI images of BIPI children; the regions of interest of the lesions in the brain MRI images have been labeled; Step S2: Preprocess the data set; Step S3: Randomly divide the preprocessed data set into a first training set and a first test set; Step S4: Build an automatic lesion segmentation model for BIPI children; wherein, the automatic lesion segmentation model for BIPI children is a first deep learning segmentation model; Step S5: Input the first training set into the segmentation model for training and optimization, and save the model parameters corresponding to the first training set; Step S6: Input the first test set into the trained segmentation model to automatically obtain the lesion segmentation results corresponding to the first test set; Step S7: Score the brain MRI images of BIPI children in the data set according to the Kidokoro scoring system standard to obtain the Kidokoro scoring results corresponding to the data set; Step S8: Build a KD grading prediction model; wherein, the KD grading prediction model is a second deep learning segmentation model; Step S9: Use the lesion segmentation results of the data set as the input of the KD grading prediction model, use the Kidokoro scoring results corresponding to the data set as the labels of the data set, and randomly divide the lesion segmentation results of the data set into a second training set and a second test set; Step S10: Input the second training set into the KD grading prediction model for training and optimization, and save the model parameters corresponding to the second training set; Step S11: Input the second test set into the trained KD grading prediction model to automatically obtain the Kidokoro scoring results corresponding to the second test set; Step S12: Build a lesion feature extraction network model; wherein, the lesion feature extraction network model is a convolutional neural network; Step S13: When inputting the lesion segmentation results of the data set into the lesion feature extraction network model, automatically extract the imaging features of the brain MRI images of BIPI children corresponding to the data set; Step S14: Obtain the perinatal clinical index data of BIPI children corresponding to the data set; Step S15: Classify and grade the perinatal clinical index data; Step S16: Obtain the GMs evaluation results corresponding to the data set; Step S17: Quantitatively convert the Kidokoro scoring results, the classified and graded perinatal clinical index data, and the GMs evaluation results corresponding to the data set respectively; Step S18: Build a cerebral palsy risk prediction model for BIPI children, and use the quantitatively converted GMs evaluation results as the output of the cerebral palsy risk prediction model for BIPI children; wherein, the cerebral palsy risk prediction model for BIPI children is a third deep learning segmentation model;Step S19: Integrate the imaging features of the brain MRI images of the BIPI children corresponding to the dataset, the quantized and converted Kidokoro score results, and the quantized and converted perinatal clinical index data into a combination, and randomly divide the combination into a third training set and a third test set; Step S20: Input the third training set into the cerebral palsy risk prediction model for BIPI children for training and optimization, and save the model parameters; Step S21: Input the third test set into the trained cerebral palsy risk prediction model for BIPI children to automatically obtain the cerebral palsy risk prediction results for the BIPI children corresponding to the third test set.

[0010] Preferably, the sequence of the brain MRI image includes one or a combination of the following: T1WI, T2WI, DWI, T2-FLAIR.

[0011] Preferably, the automatic segmentation model for the lesions of BIPI children is the Attention-UNet deep learning segmentation network; the KD grading prediction model is CNN; the lesion feature extraction network model is the three-dimensional convolutional neural network 3D-ResNet; the cerebral palsy risk prediction model for BIPI children is the XGBoost model.

[0012] Preferably, Step S7 specifically includes: Step S7.1: Automatically segment 6 structural regions in the brain MRI images corresponding to the dataset, and automatically measure the values corresponding to the 6 structural regions; wherein, the values corresponding to the 6 structural regions are respectively: the width of the bilateral parietal lobes, the interhemispheric distance, the thickness of the corpus callosum, the longest diameter of the cerebellar hemisphere, the widest diameter of the body of the left and right lateral ventricles, and the area of the basal ganglia region; Step S7.2: Based on the automatic segmentation results of the 6 structural regions and the automatic measurement results of the values corresponding to the 6 structural regions, score the brain MRI images of the BIPI children in the dataset according to the Kidokoro scoring system standard to obtain the Kidokoro score results corresponding to the dataset.

[0013] Preferably, the perinatal clinical index data includes one or a combination of the following: gestational age at birth, birth weight, gender, mode of delivery, Apgar 1-minute score, Apgar 5-minute score, etiological factors, C-reactive protein, cerebrospinal fluid test indicators.

[0014] Advantages of the present invention:

[0015] The method for predicting the risk of cerebral palsy in BIPI children provided by the present invention combines multi-dimensional factor information such as MRI imaging signs, KD grading, perinatal factors, and clinical laboratory indicators. Based on the GMs assessment results, it innovatively develops and validates a risk prediction model for early predicting the risk of cerebral palsy and developmental disorders in BIPI children, aiming to improve the accuracy and practicality of early prediction, identify the high and low risk rates of cerebral palsy earlier, provide an objective and quantifiable support basis for clinical early intervention and formulating rehabilitation plans, and also provide important help for parents of children to actively cooperate or reduce excessive anxiety.

[0016] Additional aspects and advantages of the present invention will become apparent in the following description or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The schematic flowchart of the method for predicting the risk of cerebral palsy in BIPI children according to an embodiment of the present invention is shown;

[0018] Figure 2 The schematic flowchart of constructing a KD grading prediction model based on the T1WI+T2WI combined sequence of MRI using a deep learning method according to an embodiment of the present invention is shown;

[0019] Figure 3 The schematic flowchart of constructing a KD grading prediction model for BIPI lesion features based on a deep learning method according to an embodiment of the present invention is shown;

[0020] Figure 4 The key schematic diagram of the XGBoost model extracting lesion features through deep learning according to an embodiment of the present invention is shown;

[0021] Figure 5 The schematic flowchart of constructing a cerebral palsy risk prediction model according to an embodiment of the present invention is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to more clearly understand the above objects, features, and advantages of the present invention, as Figures 1 to 5 shown, the present invention will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0023] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the limitations of the specific embodiments disclosed below.

[0024] Figure 1The schematic flowchart of the method for predicting the risk of cerebral palsy in BIPI children according to an embodiment of the present invention is shown. As Figure 1 shown, the method for predicting the risk of cerebral palsy in BIPI children includes: building an automatic lesion segmentation model for BIPI children; automatically obtaining the lesion segmentation result based on the automatic lesion segmentation model for BIPI children; building a KD grading prediction model; automatically obtaining the Kidokoro score result based on the KD grading prediction model; building a lesion feature extraction network model; automatically extracting the imaging features of the brain MRI images of BIPI children based on the lesion feature extraction network model; building a risk prediction model for cerebral palsy in BIPI children, and taking the GMs evaluation result after quantitative conversion as the output of the risk prediction model for cerebral palsy in BIPI children; integrating the imaging features of the brain MRI images of BIPI children corresponding to the data set, the Kidokoro score result after quantitative conversion, and the perinatal clinical index data after quantitative conversion into a combination, and taking this combination as the input of the risk prediction model for cerebral palsy in BIPI children; automatically obtaining the risk prediction result of cerebral palsy in BIPI children.

[0025] In this embodiment, the method for predicting the risk of cerebral palsy in BIPI children provided by the present invention combines multi-dimensional factor information such as MRI imaging signs, KD grading, perinatal factors, and clinical laboratory indicators, and based on the GMs evaluation result, innovatively develops and validates a risk prediction model for early predicting cerebral palsy and developmental disorders in BIPI children, aiming to improve the accuracy and practicality of early prediction, identify the high and low risk rates of cerebral palsy earlier, provide an objective and quantifiable support basis for clinical early intervention and formulating rehabilitation plans, and also provide important help for parents of children to actively cooperate or reduce excessive anxiety.

[0026] In one embodiment of the present invention, a method for predicting the risk of cerebral palsy in BIPI children specifically includes: Step S1: Obtain a data set; wherein, the data set is the brain MRI images of BIPI children; the regions of interest of the lesions in the brain MRI images have all been labeled; Step S2: Preprocess the data set; Step S3: Randomly divide the preprocessed data set into a first training set and a first test set; Step S4: Build an automatic lesion segmentation model for BIPI children; wherein, the automatic lesion segmentation model for BIPI children is a first deep learning segmentation model; Step S5: Input the first training set into the segmentation model for training and optimization, and save the model parameters corresponding to the first training set; Step S6: Input the first test set into the trained segmentation model to automatically obtain the lesion segmentation results corresponding to the first test set; Step S7: Score the brain MRI images of BIPI children in the data set according to the Kidokoro scoring system standard to obtain the Kidokoro scoring results corresponding to the data set; Step S8: Build a KD grading prediction model; wherein, the KD grading prediction model is a second deep learning segmentation model; Step S9: Use the lesion segmentation results of the data set as the input of the KD grading prediction model, use the Kidokoro scoring results corresponding to the data set as the labels of the data set, and randomly divide the lesion segmentation results of the data set into a second training set and a second test set; Step S10: Input the second training set into the KD grading prediction model for training and optimization, and save the model parameters corresponding to the second training set; Step S11: Input the second test set into the trained KD grading prediction model to automatically obtain the Kidokoro scoring results corresponding to the second test set; Step S12: Build a lesion feature extraction network model; wherein, the lesion feature extraction network model is a convolutional neural network; Step S13: When inputting the lesion segmentation results of the data set into the lesion feature extraction network model, automatically extract the image features of the brain MRI images of BIPI children corresponding to the data set; Step S14: Obtain the perinatal clinical index data of BIPI children corresponding to the data set; Step S15: Classify and grade the perinatal clinical index data; Step S16: Obtain the GMs assessment results corresponding to the data set; Step S17: Quantitatively transform the Kidokoro scoring results, the classified and graded perinatal clinical index data, and the GMs assessment results corresponding to the data set respectively; Step S18: Build a cerebral palsy risk prediction model for BIPI children; wherein, the cerebral palsy risk prediction model for BIPI children is a third deep learning segmentation model;Step S19: Integrate the imaging features of the brain MRI images of BIPI children corresponding to the dataset, the quantized and converted Kidokoro score results, the quantized and converted perinatal clinical index data, and the quantized and converted GMs assessment results into a combination, and randomly divide this combination into a third training set and a third test set; Step S20: Input the third training set into the BIPI children's cerebral palsy risk prediction model for training and optimization, and save the model parameters; Step S21: Input the third test set into the trained BIPI children's cerebral palsy risk prediction model to automatically obtain the BIPI children's cerebral palsy risk prediction results corresponding to the third test set.

[0027] In one embodiment of the present invention, the sequence of the brain MRI image includes one or a combination of the following: T1WI, T2WI, DWI, T2-FLAIR.

[0028] In one embodiment of the present invention, the automatic segmentation model for the lesions of BIPI children is the Attention-UNet deep learning segmentation network; the KD grading prediction model is CNN; the lesion feature extraction network model is the three-dimensional convolutional neural network 3D-ResNet; the BIPI children's cerebral palsy risk prediction model is the XGBoost model.

[0029] In one embodiment of the present invention, step S7 specifically includes: Step S7.1: Automatically segment 6 structural regions in the brain MRI images corresponding to the dataset, and automatically measure the values corresponding to the 6 structural regions; wherein, the values corresponding to the 6 structural regions are respectively: the width of the bilateral parietal lobes, the interhemispheric distance, the thickness of the corpus callosum, the longest diameter of the cerebellar hemisphere, the widest diameter of the body of the left and right lateral ventricles, and the area of the basal ganglia region; Step S7.2: Based on the automatic segmentation results of the 6 structural regions and the automatic measurement results of the values corresponding to the 6 structural regions, score the brain MRI images of BIPI children in the dataset according to the Kidokoro scoring system standard to obtain the Kidokoro score results corresponding to the dataset.

[0030] In one embodiment of the present invention, the perinatal clinical index data includes one or a combination of the following: gestational age at birth, birth weight, gender, mode of delivery, APgar 1min score, APgar 5min score, causative etiology, C-reactive protein, cerebrospinal fluid test indicators.

[0031] In this embodiment, the clinical GMs assessment results and the multi-modal combined sequence lesion features determine that the BIPI children's cerebral palsy risk prediction model is characterized by incorporating multi-dimensional factors for comprehensive evaluation.

[0032] In one embodiment of the present invention, asFigure 5 As shown, the process of constructing the cerebral palsy risk prediction model of the present invention specifically includes: using the GMS score as a label and multimodal lesions as inputs to train the ResNet network, and at the same time converting other features (such as etiology classification, etc.) into integer labels as clinical features; after using the GMS score as a label and multimodal lesions as inputs to train the ResNet network, after fine-tuning, the network feature extraction part is separately separated; using the feature extractor to extract the lesion features in the MRI as deep learning features; combining the MRI deep learning features and clinical prognosis labels of the cases, and dividing the samples into a training set, a test set and a validation set; using the MRI deep learning features and clinical features of the samples as inputs, and high-risk and low-risk clinical prognosis as labels, to construct an XGBoost regression model; using the network search method to search for the optimal hyperparameters, and using this set of hyperparameters to train the XGBoost model; after training is completed, test in the test set to measure the prediction effect of the high-risk and low-risk classification model of clinical prognosis.

[0033] The following will show the technical solution of the present invention with a specific embodiment. The specific implementation steps of a method for predicting the cerebral palsy risk of BIPI children according to the present invention are as follows:

[0034] (1) Automatically obtain the lesion segmentation result. Specifically, it includes the following steps:

[0035] Step S1: Obtain a dataset; select 507 children diagnosed with BIPI in the Neonatology Department of the Children's Hospital Area, Obstetrics and Gynecology and Women and Children's Hospital Area of Dalian Women and Children's Medical Center (Group) from March 2019 to March 2023 as the research objects, with a gestational age of 24 to 36 weeks; among them, 307 cases in the Children's Hospital Area are used as the main data for model construction. Among them, the dataset is the brain MRI images of BIPI children; the sequences of the brain MRI images include one or a combination of the following: T1WI, T2WI, DWI, T2-FLAIR. T1WI is T1 weighted imaging (abbreviated as T1WI), T2WI is T2 weighted imaging (abbreviated as T2WI). DWI is Diffusion weighted imaging (abbreviated as DWI). T2-FLAIR is T2-Fluid Attenuated Inversion Recovery sequence (abbreviated as T2-FLAIR). The regions of interest of the lesions in the brain MRI images have all been labeled; Step S2: Preprocess the dataset; Step S3: Randomly divide the preprocessed dataset into a first training set and a first test set; Step S4: Build an automatic lesion segmentation model for BIPI children; among them, the automatic lesion segmentation model for BIPI children is the first deep learning segmentation model; the automatic lesion segmentation model for BIPI children is the Attention-UNet deep learning segmentation network; Step S5: Input the first training set into the segmentation model for training and optimization, and save the model parameters corresponding to the first training set; Step S6: Input the first test set into the trained segmentation model to automatically obtain the lesion segmentation results corresponding to the first test set.

[0036] In the process of automatically obtaining the lesion segmentation results, first, the brain MRI image dataset of BIPI children is acquired, preprocessed, and the preprocessed dataset is randomly divided into a first training set and a first test set. Then, a lesion automatic segmentation model for BIPI children is built. The first training set is input into the segmentation model for training and optimization, and the model parameters corresponding to the first training set are saved. The first test set is input into the trained segmentation model to automatically obtain the lesion segmentation results corresponding to the first test set, achieving the purpose of early diagnosis and segmentation of the lesions of BIPI children using deep learning methods and based on conventional MRI sequences. Specifically, high-dimensional information features that cannot be recognized by the human eye are mined through deep learning technology to enhance the clinical application value of conventional MRI sequences and also improve the primary clinical applicability of the built lesion automatic segmentation model for BIPI children, providing fast and reliable intelligent auxiliary diagnosis for BIPI children and laying a good foundation for multiple problems such as improving the diagnostic accuracy of imaging diagnosticians, reducing the diagnostic time, and the examination burden of BIPI children.

[0037] (2) Automatically obtain the Kidokoro score results. The specific steps are as follows:

[0038] Step S7: Score the brain MRI images of the BIPI children in the dataset according to the Kidokoro scoring system standard to obtain the Kidokoro score results corresponding to the dataset; Step S8: Construct a KD grading prediction model; where the KD grading prediction model is a second deep learning segmentation model; Step S9: Use the lesion segmentation results of the dataset as the input of the KD grading prediction model, use the Kidokoro score results corresponding to the dataset as the label of the dataset, and randomly divide the lesion segmentation results of the dataset into a second training set and a second test set; Step S10: Input the second training set into the KD grading prediction model for training and optimization, and save the model parameters corresponding to the second training set; Step S11: Input the second test set into the trained KD grading prediction model to automatically obtain the Kidokoro score results corresponding to the second test set.

[0039] Step S7 specifically includes: Step S7.1: Automatically segment 6 structural regions in the brain MRI images corresponding to the data set, and automatically measure the values corresponding to the 6 structural regions; among them, the values corresponding to the 6 structural regions are: the width of the bilateral parietal lobes, the interhemispheric distance, the thickness of the corpus callosum, the longest diameter of the cerebellar hemisphere, the widest diameter of the body parts of the left and right lateral ventricles, and the area of the basal ganglia region; Step S7.2: Based on the automatic segmentation results of the 6 structural regions and the automatic measurement results of the values corresponding to the 6 structural regions, score the brain MRI images of the BIPI children in the data set according to the Kidokoro scoring system standard to obtain the Kidokoro scoring results corresponding to the data set.

[0040] In this embodiment, before automatically segmenting the 6 structural regions in the brain MRI images corresponding to the data set and automatically measuring the values corresponding to the 6 structural regions, the measurement values of the 6 important structural regions need to be manually measured. The values corresponding to the 6 structural regions are: the width of the bilateral parietal lobes (FLV), the interhemispheric distance (IF), the thicknesses of 3 positions (genu, body, splenium) of the corpus callosum (CC), the longest diameter of the cerebellar hemisphere (CV), the widest diameter of the body parts of the left and right lateral ventricles (LV), and the area of the basal ganglia region (TCV) (Table 1).

[0041] Table 1 Manual measurement methods for distances and areas of six important structural regions

[0042]

[0043] The measurements are performed by 2 deputy senior physicians with 10 years of pediatric MRI diagnosis experience. First, Physician 1 performs manual measurements to obtain the values, and then Physician 2 checks the measurement sites and measures again to obtain the values. The average of the final measurement results of Physician 1 and Physician 2 is taken as the final result of the measurement value.

[0044] Furthermore, automatically segment the 6 structural regions in the brain MRI images corresponding to the data set, and automatically measure the values corresponding to the 6 structural regions. In a specific embodiment, next we will describe the methods for automatic segmentation and automatic measurement of the six structural regions. Label the 6 structural regions in the brain MRI images corresponding to the data set respectively.

[0045] Use the 3D-Unet network for segmentation. The specific segmentation and automatic measurement methods are as follows:

[0046] Segmentation method: First, classify the 6 structural region annotations corresponding to the original image, and perform data augmentation using intensity normalization, removing the black background, rotation, and scaling respectively. Input the small blocks randomly cropped to (64, 64, 32) into 3D-UNET, obtain the feature map after downsampling using maxpooling, use 3D convolution, batchnorm3D, and RELU layers to form the feature extraction layer, use transposed convolution for upsampling, and pass through skip connections in the middle to provide high-resolution features for the decoder. The last layer uses 1x1x1 convolution to adjust the number of channels to obtain the segmentation result. Calculate the loss between the segmentation result and the annotation, and backpropagate the gradient to adjust the network parameters. After training converges, finally perform the segmentation of the six structures on the test set.

[0047] Automatic measurement: We wrote an automatic measurement program using the shapely library in Python. This program can construct computational geometric models of different brain regions and automatically calculate the sizes of these regions. First, preprocess the T2WI and T1WI images required for measurement using slice interpolation (upsampling to a resolution of 0.7x0.7x1mm3) and N4 bias field correction. Use the FMRIB linear image registration tool (FMRIB PPREMO population template), and use the obtained transformation to transform the segmentation into this space while keeping the size and shape consistent. Among them, since the corpus callosum is a relatively thin region, select the 97th percentile to determine the distance at the thickest position of each part of the corpus callosum respectively, reflecting a position similar to manual measurement. The automatic measurement process of the 6 structural regions is as follows:

[0048] IF measurement: Use the automatic measurement program to collect all the boundary points in the IF structure, traverse and calculate the distance between two points with the same y coordinate, and then compare and select the maximum value as the interhemispheric distance (IF).

[0049] LV measurement: Use the automatic measurement program to collect all the boundary points in the LV structure, traverse all the points on the left and right sides of a single LV structure respectively, measure the left and right diameters, and calculate the widest diameter of the left LV as the widest diameter of the body of the left lateral ventricle, and the widest diameter of the right LV as the widest diameter of the body of the right lateral ventricle.

[0050] CV measurement: Use the automatic measurement program to collect all the boundary points in the cerebellar hemisphere, traverse all the points on the left and right sides respectively, and calculate the maximum distance from the left boundary to the right boundary of the image as the longest diameter of the cerebellar hemisphere (CV). FLV measurement: Similar to the CV measurement, use the automatic measurement program to collect all the boundary points in the FLV, and calculate the maximum distance from the left boundary to the right boundary of the image as the biparietal width (FLV).

[0051] CC measurement: All boundary points of the corpus callosum are collected using an automatic measurement program. The central point A of the medial geniculate body is obtained through coordinate comparison, the lowermost boundary point B near the body, and the lowermost boundary point C near the splenium. The body, genu, and splenium of the CC are demarcated by these three points. By traversing the medial boundary points, the perpendicular lines corresponding to the medial and lateral boundary points are found and intersect at the lateral point, the distance between the two points is calculated, and the thicknesses of the body, genu, and splenium are obtained by comparing the distances.

[0052] Further, based on the automatic segmentation results of 6 structural regions and the automatic measurement results of the corresponding numerical values of the 6 structural regions, the brain MRI images of the BIPI children in the dataset are scored according to the Kidokoro scoring system criteria to obtain the Kidokoro scoring results corresponding to the dataset. In a specific embodiment, next, we will describe the scoring content and scoring criteria of the Kidokoro scoring system.

[0053] The MRI images of all enrolled BIPI children are scored according to the Kidokoro scoring system criteria. The specific scoring items and result judgment criteria are shown in Tables 2.1 - 2.4:

[0054] The items of the Kidokoro scoring system are divided into 4 major categories: white matter abnormalities, cortical gray matter (GM) abnormalities, deep gray matter abnormalities, and cerebellar abnormalities. Among them, the evaluation content and scoring rules for white matter abnormalities are shown in Table 2.1. The evaluation content and scoring rules for cortical gray matter abnormalities are shown in Table 2.2. The evaluation content and scoring rules for deep gray matter abnormalities are shown in Table 2.3. The evaluation content and scoring rules for cerebellar structural abnormalities are shown in Table 2.4.

[0055] Table 2.1 Evaluation content and scoring rules for white matter abnormalities

[0056]

[0057] Note: FLV represents the maximum width of the bilateral parietal lobes in the coronal plane; LV represents the width of the body of the lateral ventricle.

[0058] Table 2.2 Evaluation content and scoring rules for cortical gray matter abnormalities

[0059]

[0060] Note: IF represents the maximum width of the interhemispheric distance measured in a single-layer axial plane of the largest cross-section of the bilateral superior frontal gyri

[0061] Table 2.3 Evaluation content and scoring rules for deep gray matter abnormalities

[0062]

[0063]

[0064] Note: TCV represents the maximum cross-sectional area of the basal ganglia measured on the axial plane where the maximum visibility is at the caudate head and lentiform nucleus.

[0065] Table 2.4 Content and Scoring Rules for the Evaluation of Cerebellar Structural Abnormalities

[0066]

[0067] Note: CV represents the maximum diameter value of the bilateral cerebellar hemispheres on the axial image.

[0068] The total score of brain WM abnormalities is divided into 4 grades: none (total score 0 - 2), mild (total score 3 - 4), moderate (total score 5 - 6), severe (total score > 7).

[0069] The total score of cortical GM abnormalities is divided into 4 grades: none (total score 0), mild (total score 1), moderate (total score 2), and severe (total score > 3);

[0070] The deep GM abnormalities and cerebellar abnormalities are each divided into 4 grades: none (total score 0), mild (total score 1), moderate (total score 2), and severe (total score > 3);

[0071] The total score of overall brain abnormalities is divided into 4 grades: Grade 0: normal (total score 0 - 3), Grade 1: mild (total score 4 - 7), Grade II: moderate (total score 8 - 11), and Grade III: severe (total score > 12).

[0072] Furthermore, for the brain MRI images of BIPI children, two deputy senior physicians with over 10 years of rich neonatal MRI diagnosis work experience and familiar with the KD scoring rules scored according to the above MRI scoring system's specified item content and rules. The scoring of each item was independently carried out by 2 experts, and the final scoring result value was based on the average of the scores of the 2 experts. Each child finally obtained the total scores of four single items: the total score of brain WM, the total score of cortical GM, the total score of deep GM, and the total score of the cerebellum. Then, the total scores of the 4 items were aggregated to obtain the total score of overall brain abnormalities for each child. This value was graded according to the scoring standard, and finally, the grading level of each child was obtained. This is for subsequent use of the corresponding Kidokoro scoring results of the dataset as the label of the input dataset of the KD grading prediction model.

[0073] In a specific embodiment, the KD grading prediction model is 3D-ResNet in CNN, as Figure 2 and Figure 3As shown, the size of the input image is (128, 128, 32). 3D-ResNet contains multiple residual blocks. Each residual block (ResidualBlock) usually contains multiple convolutional layers, pooling layers, and batch normalization layers. The first convolutional layer of 3D-ResNet is used to process the input data, usually with a relatively large convolutional kernel, such as a 3x7x7 convolution, and a stride of 1. Inside the residual block, there are usually multiple convolutional layers, and these convolutional layers usually have relatively small convolutional kernels, such as a 3x3x3 convolution. After the convolutional layer, there will be a batch normalization layer (Batch Normalization), which helps to accelerate training and helps the model converge. Inside the residual block, the RELU activation function is often used to add non-linearity to the model. The residual block contains skip connections, which are mainly used to solve the problems of gradient disappearance and gradient explosion during the training process of deep neural networks, making the deep neural network easier to train. The skip connection adds the input of the residual block to the output of the last convolutional layer. If the input size and output size are different, the size is adjusted through a 1x1x1 convolution. After the residual block, a pooling layer is used to reduce the size of the feature map. After multiple residual blocks are stacked, the network uses a fully connected layer for classification.

[0074] During the training process, the cross-entropy loss function is used. The cross-entropy loss function is defined as:

[0075]

[0076] In the formula, C is the total number of categories, y ic is an indicator function that is 1 when sample i belongs to category c and 0 otherwise. pic is the probability that the model predicts that sample i belongs to category c. During the training process, the Adam optimizer is used, and the initial learning rate is set to 0.001. The learning rate adjustment method uses the cosine annealing learning rate adjustment method. The learning rate is relatively large at the beginning of training and gradually decreases during the later stage of training, which helps to converge the model in the early stage and finely adjust the model parameters in the later stage to improve the training effect of the model. The model is trained on the training set data, and the performance on the validation set is monitored to adjust the hyperparameters. The accuracy, recall rate, F1 score, etc. of the model are evaluated on the test set.

[0077] Using the T1WI single sequence and the T1WI+T2WI combined sequence lesions that have been segmented in the early stage, 307 cases were randomly grouped according to the ratio of 8:1:1. The deep learning network framework selected was 3D-ResNet in CNN, the input image size was (128, 128, 32), 3D-ResNet contained multiple residual blocks, and a KD grading prediction model for preterm brain injury was constructed. Moreover, the deep learning method has more advantages in constructing the KD prediction model for BIPI injury level, and the lesion characteristics of the MRI combined T1WI+T2WI sequence perform best in model construction, demonstrating the potential of AI technology in improving the diagnostic efficiency of MRI and reducing the scanning cost of children.

[0078] (3) Automatically extract the imaging features of the brain MRI images of BIPI children corresponding to the dataset. The specific steps are as follows:

[0079] Step S12: Construct a lesion feature extraction network model; among them, the lesion feature extraction network model is a convolutional neural network; the lesion feature extraction network model is a three-dimensional convolutional neural network 3D-ResNet; Step S13: When the lesion segmentation result of the dataset is input into the lesion feature extraction network model, automatically extract the imaging features of the brain MRI images of BIPI children corresponding to the dataset.

[0080] (4) Step S14: Obtain the perinatal clinical index data of BIPI children corresponding to the dataset.

[0081] The perinatal clinical index data includes one or a combination of the following: gestational age at birth, birth weight, gender, mode of delivery, Apgar 1-minute score, Apgar 5-minute score, causative etiology, C-reactive protein, cerebrospinal fluid test indicators. The clinical GMs assessment results and multi-modal combined sequence lesion characteristics determine that the characteristics of the cerebral palsy risk prediction model for BIPI children are that multi-dimensional factors are included for comprehensive evaluation.

[0082] (5) Step S15: Classify and grade the perinatal clinical index data.

[0083] Classifying and grading the perinatal clinical index data specifically includes: classifying and grading each clinical index according to the standards of "Expert Consensus on the Diagnosis and Prevention of Preterm Brain Injury", "Practical Prematurity and Preterm Infants", and "Practical Neonatology" as follows:

[0084] Gender: Divided into 2 categories: male and female.

[0085] Birth weight: Premature infants were divided into 4 groups according to birth weight. Extremely low birth weight group (ELBWI): birth weight ≤ 1000 g; Very low birth weight group (VLBW): birth weight ≤ 1500 g; Low birth weight group (LBW): 1500 < birth weight < 2500 g; Normal birth weight group (NBW): birth weight ≥ 2500 g.

[0086] Gestational age at birth: Gestational age at birth was divided into 4 groups: Extremely premature infants group: gestational age at birth ≤ 28 weeks; Early premature infants group: 28 weeks < gestational age at birth ≤ 31 weeks; Middle premature infants group: 32 weeks ≤ gestational age at birth < 35 weeks; Late premature infants group: 35 weeks ≤ gestational age at birth < 37 weeks.

[0087] Mode of delivery: Divided into 2 categories: vaginal delivery and cesarean section.

[0088] Apgar score: The Apgar scores were evaluated at 2 time periods and graded as follows:

[0089] 1 min (for asphyxia diagnosis and classification) Grade I: 0 - 3 points; Grade II: 4 - 7 points; Grade III: 8 - 10 points.

[0090] 5 min (for determining the resuscitation effect and prognosis) Grade I: 0 - 3 points; Grade II: 4 - 7 points; Grade III: 8 - 10 points.

[0091] Etiological classification: The etiologies leading to BIPI were divided into 4 major categories in total: Category A (hypoxic - ischemic encephalopathy and hemodynamic disorders): including severe intrauterine distress and asphyxia at birth, hyper - or hypo - capnia, circulatory failure / shock, hypotension / hypertension / abnormal blood pressure fluctuations, heart failure, respiratory failure, mechanical ventilation, severe dehydration, hypothermia, intrauterine growth restriction, severe or complex congenital heart diseases; Category B (infection and inflammatory response): including chorioamnionitis, intrauterine infection, postnatal infection, necrotizing enterocolitis, etc.; Category C (hematological diseases): including coagulation abnormalities, moderate - to - severe anemia, polycythemia - hyperviscosity syndrome, antithrombin deficiency, plasminogen deficiency; Category D (obstetric risk factors): including thrombus or amniotic fluid embolism, maternal complications / comorbidities and bad habits (hypertension, heart disease, diabetes, severe anemia, smoking, drug use), abnormal delivery history (emergency cesarean section, placental abruption, forceps / vacuum extraction, shoulder dystocia, precipitate labor, prolonged labor).

[0092] C - reactive protein (CRP): Group A (normal) 0 - 15 mg / L, Group B (abnormal) > 15 mg / L.

[0093] Cerebrospinal fluid test (ADA, sugar, LDH): divided into 2 groups, Group A (normal) ADA <40 (U / L), LDH <240U / L, sugar <4.5mmol / L, Group B (abnormal) ADA >40 (U / L), LDH >240U / L, sugar >4.5mmol / L.

[0094] (6) Step S16: Obtain the GMs evaluation result corresponding to the data set.

[0095] Specifically, the children corresponding to the data set were evaluated for motor development by professional GMs assessors in the rehabilitation center. Furthermore, a standardized video assessment method was used; the first twisting stage was recorded once from 3 days after birth to full term, and the second restless twisting stage was recorded once from 2 months to 5 months after full term. For those who were evaluated as spasticity-synchronicity (CS) and restless movement deficiency (F-) in the first two stages, they were evaluated again after 12 months of full term. Furthermore, two rehabilitation professionals who obtained the GM Trust training course qualification certificate and had assessment qualifications participated in the assessment. Each time, one assessor (therapist) conducted the initial assessment first, and then another more experienced assessor (rehabilitation therapist) conducted the reassessment. The assessors used visual Gestalt perception to evaluate each GMs performance and distinguished normal and abnormal subclassifications for different stages. The auditory signal of the video was turned off during the assessment. When more abnormal records were assessed or difficulties occurred during the assessment, the perception was recalibrated using standard normal GMs records. It takes about 5 minutes to evaluate a single recording.

[0096] The results of GMs are described below. Stage 1 (twisting stage): The evaluation results are divided into normal and abnormal. Normal results are: normal twisting (N); abnormal results are: monotonous whole body movement (PR), spastic-synchronous whole body movement (CS), chaotic whole body movement (Ch). PR refers to the monotonous sequence of each continuous movement component, and the movement of different parts of the body has lost its normal complexity. CS refers to rigid movement, loss of normal fluidity, and almost simultaneous contraction and relaxation of all limbs and trunk muscles. Ch refers to large amplitude of movement of all limbs, chaotic sequence, loss of fluidity, and sudden and incoherent movements.

[0097] Stage 2 (restlessness stage): The assessment results are divided into normal and abnormal. Normal is: normal restless movement (NF), abnormal is: lack of restless movement (F-), abnormal restless movement (AF). NF is a small-amplitude, medium-speed movement that occurs throughout the neck, trunk and limbs in all directions with variable movement acceleration. In awake infants, this movement persists except when they are fussy and crying, and can exist at the same time as other movements. The frequency of restless movements changes with age. F- refers to the fact that restless movements have not been observed within weeks to months after full term, which is called "lack of restless movements." AF is moderately or significantly exaggerated in terms of movement amplitude, speed and unsteadiness.

[0098] Those with normal first-stage assessment can be determined to have good development. Those with abnormal assessment need to be followed up until the second stage. Those with normal results in the second stage are determined to have good development, and those with abnormal results indicate cerebral palsy and motor developmental delay. Among them, F-tendency indicates cerebral palsy, and AF-tendency indicates motor developmental delay. Finally, the result of the second stage is taken as the final judgment outcome, and the GMs results are dichotomized into high- and low-risk groups. The low-risk group includes those with normal results (N, NF), and the high-risk group includes those with abnormal results (F-, AF).

[0099] (7) Step S17: Quantitatively convert the Kidokoro score results, the perinatal clinical index data after classification and grading, and the GMs assessment results corresponding to the data set respectively.

[0100] Quantitatively convert the Kidokoro score results, the perinatal clinical index data after classification and grading, and the GMs assessment results corresponding to the data set respectively, specifically including:

[0101] Collect 9 types of perinatal clinical index data of BIPI children, including gestational age at birth, birth weight, gender, mode of delivery, Apgar 1-minute score, Apgar 5-minute score, causative etiology, C-reactive protein (CRP), cerebrospinal fluid test (ADA, sugar, LDH). Classify and organize the 9 types of perinatal clinical index data according to the previous research classification criteria for subsequent conversion of these indicators into positive integers as independent variables for the preliminary input of the XGBoost model. The four-category classification of gestational age at birth (A, B, C, D) is converted into positive integers 0, 1, 2, 3; the four-category classification of birth weight (A, B, C, D) is converted into positive integers 0, 1, 2, 3; the two-category classification of gender (A, B) is converted into positive integers 0, 1; the two-category classification of mode of delivery (A, B) is converted into positive integers 0, 1; the three-category classification of Apgar (1-minute and 5-minute) scores (A, B, C) is converted into positive integers 0, 1, 2; the four-category classification of the main etiology (A, B, C, D) is converted into positive integers 0, 1, 2, 3; the two-category classification of C-reactive protein (CRP) (A, B) is converted into positive integers 0, 1; the two-category classification of cerebrospinal fluid test (A, B) is converted into positive integers 0, 1. The four-category conversion of the KD classification of the degree of whole-brain abnormality of the Kidokoro score results corresponding to the data set (normal, mild injury, moderate injury, severe injury) is converted into positive integers 0, 1, 2, 3. Convert the GMS prognosis classification of the GMs assessment results (low risk, high risk) into positive integer labels of 0 and 1 as the final output low- and high-risk labels.

[0102] (8) Automatically obtain the cerebral palsy risk prediction results of BIPI children. Specifically, it includes the following steps:

[0103] Step S18: Construct a cerebral palsy risk prediction model for BIPI children, and use the quantitatively converted GMs evaluation results as the output of the cerebral palsy risk prediction model for BIPI children; wherein, the cerebral palsy risk prediction model for BIPI children is the third deep learning segmentation model; the cerebral palsy risk prediction model for BIPI children is an XGBoost model; Step S19: Integrate the imaging features of the brain MRI images of the BIPI children corresponding to the dataset, the quantitatively converted Kidokoro score results, and the quantitatively converted perinatal clinical index data into a combination, and randomly divide this combination into a third training set and a third test set; Step S20: Input the third training set into the cerebral palsy risk prediction model for BIPI children for training and optimization, and save the model parameters; Step S21: Input the third test set into the trained cerebral palsy risk prediction model for BIPI children, and automatically obtain the cerebral palsy risk prediction results of the BIPI children corresponding to the third test set.

[0104] Table 3 Construction results of the XGBoost model for multimodal cerebral palsy risk prediction of each combined sequence

[0105]

[0106] Table 3 shows the construction results of the XGBoost model for multimodal cerebral palsy risk prediction of each combined sequence. It can be known from Table 3 that the XGBoost model for multimodal cerebral palsy risk prediction constructed by the T1WI+T2WI sequence combination is the optimal model.

[0107] After taking the optimally segmented lesions of the T1WI+T2WI combined sequence that had been segmented in the previous study as the input, we adopted a key strategy: freeze the parameters of the model classification layer (fully connected layer), continue to fine-tune on the training set, and after convergence, carefully separate the feature extraction part of the model, including multiple layers of convolution, downsampling pooling, and BN (Batch Normalization) layers - that is, intercept the fully connected layer of the model, which is the part to the left of the FC in the figure (the part to the left of the dotted line), as Figure 4 shown. After separation, a separate feature extractor part is obtained, which we save as a model. This part can extract the imaging features of the image while inputting the MRI lesions. We used this feature extractor for 307 segmented lesions, extracted the feature maps, and saved the extracted feature map matrices as.npy files for use in the XGboost (Extreme Gradient Boosting) model. In this process, the features of the lesions are efficiently extracted by the deep learning feature extractor and used as one of the key features in XGBoost.

[0108] Table 4 Results of ablation experiment parameters of the optimal combination T1WI+T2WI sequence cerebral palsy risk prediction model

[0109]

[0110]

[0111] Table 4 shows the results of ablation experiment parameters of the optimal combination T1WI+T2WI sequence cerebral palsy risk prediction model. It can be known from Table 4 that the cerebral palsy prediction model constructed by multi-parameters of imaging signs+clinical+KD features has the best performance.

[0112] Table 5 shows the results of ablation experiment parameters of the optimal combination T1WI+T2WI sequence cerebral palsy risk prediction model. It can be known from Table 5 that after single-fusion element ablation in turn during the ablation experiment, the model performance decreases. After continuous double-fusion element ablation, the model performance gradually decreases. The Delong test of the model after each fusion element has statistical differences from the initial model, indicating that the evaluation of cerebral palsy prognosis by single-dimensional indicators is defective, and multi-dimensional indicators should be combined. Moreover, the Z-score values of the models constructed by the remaining single clinical and single MRI imaging signs are relatively small, which are 2.71 and 3.23 respectively, indicating that clinical index features have the greatest impact on the model, followed by MRI imaging signs, which are the key indicators for BIPI cerebral palsy risk prediction.

[0113] Table 5 Delong test results of each factor variable constructing the model after XGboost model ablation

[0114]

[0115] To sum up, the method for predicting the risk of cerebral palsy in BIPI children provided by the present invention combines multi-dimensional factor information such as MRI imaging signs, KD grading, perinatal factors, and clinical laboratory indicators, and based on the GMs evaluation results, innovatively develops and validates a risk prediction model for early predicting the risk of cerebral palsy and developmental disorders in BIPI children, aiming to improve the accuracy and practicability of early prediction, identify the high and low risk rates of cerebral palsy earlier, provide an objective and quantifiable support basis for clinical early intervention and formulation of rehabilitation programs, and also provide important help for parents of children to actively cooperate or reduce excessive anxiety.

[0116] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the risk of cerebral palsy in children with BIPI, characterized in that: include: Step S1: Acquire a data set; wherein the data set is a brain MRI image of a child with BIPI; the lesion regions of interest of the brain MRI image have been marked; Step S2: preprocessing the data set; Step S3: randomly divide the preprocessed data set into a first training set and a first test set; Step S4: constructing an automatic segmentation model of lesions in children with BIPI; wherein the automatic segmentation model of lesions in children with BIPI is a first deep learning segmentation model; Step S5: inputting the first training set into the segmentation model for training and optimization, and saving the model parameters corresponding to the first training set; Step S6: inputting the first test set into the trained segmentation model to automatically obtain the lesion segmentation result corresponding to the first test set; Step S7: scoring the brain MRI images of the BIPI children in the data set according to the Kidokoro scoring system standard to obtain the Kidokoro scoring result corresponding to the data set; Step S8: constructing a KD grading prediction model; wherein the KD grading prediction model is a second deep learning segmentation model; Step S9: using the lesion segmentation result of the data set as the input of the KD grading prediction model, using the Kidokoro score result corresponding to the data set as the label of the data set, and randomly dividing the lesion segmentation result of the data set into a second training set and a second test set; Step S10: inputting the second training set into the KD classification prediction model for training and optimization, and saving the model parameters corresponding to the second training set; Step S11: inputting the second test set into the trained KD grading prediction model to automatically obtain the Kidokoro score result corresponding to the second test set; Step S12: constructing a lesion feature extraction network model; wherein the lesion feature extraction network model is a convolutional neural network; Step S13: when the lesion segmentation result of the data set is input into the lesion feature extraction network model, the image features of the brain MRI image of the BIPI child corresponding to the data set are automatically extracted; Step S14: Acquire the perinatal clinical indicator data of the BIPI infant corresponding to the data set; Step S15: classifying and grading the perinatal clinical indicator data; Step S16: Obtain the GMs evaluation result corresponding to the data set; Step S17: performing quantitative conversion on the Kidokoro score results, the classified and graded perinatal clinical index data, and the GMs evaluation results corresponding to the data set respectively; Step S18: constructing a BIPI children's cerebral palsy risk prediction model, and using the quantitatively converted GMs evaluation results as the output of the BIPI children's cerebral palsy risk prediction model; wherein the BIPI children's cerebral palsy risk prediction model is a third deep learning segmentation model; Step S19: Integrate the image features of the brain MRI images of the BIPI children corresponding to the data set, the quantitatively converted Kidokoro score results, and the quantitatively converted perinatal clinical index data into a combination, and randomly divide the combination into a third training set and a third test set; Step S20: inputting the third training set into the BIPI children's cerebral palsy risk prediction model for training and optimization, and saving the model parameters; Step S21: input the third test set into the trained BIPI children's cerebral palsy risk prediction model to automatically obtain the BIPI children's cerebral palsy risk prediction result corresponding to the third test set.

2. The method for predicting the risk of cerebral palsy in children with BIPI according to claim 1, characterized in that: The sequence of brain MRI images includes one of the following or a combination thereof: T1WI, T2WI, DWI, and T2-FLAIR.

3. The method for predicting the risk of cerebral palsy in children with BIPI according to claim 1, characterized in that: The automatic segmentation model for lesions in children with BIPI is the Attention-UNet deep learning segmentation network; The KD classification prediction model is CNN; The lesion feature extraction network model is a three-dimensional convolutional neural network 3D-ResNet; The cerebral palsy risk prediction model for children with BIPI is the XGBoost model.

4. The method for predicting the risk of cerebral palsy in children with BIPI according to claim 1, characterized in that: The step S7 specifically includes: Step S7.1: Automatically segmenting the six structural regions in the brain MRI image corresponding to the data set, and automatically measuring the values ​​corresponding to the six structural regions; wherein the values ​​corresponding to the six structural regions are: biparietal lobe width, interhemispheric distance, corpus callosum thickness, longest diameter of cerebellar hemisphere, widest diameter of left and right lateral ventricle bodies, and basal ganglia area; Step S7.2: Based on the automatic segmentation results of the six structural areas and the automatic measurement results of the corresponding numerical values ​​of the six structural areas, the brain MRI images of the BIPI children in the data set are scored according to the Kidokoro scoring system standard to obtain the Kidokoro scoring results corresponding to the data set.

5. The method for predicting the risk of cerebral palsy in children with BIPI according to claim 4, characterized in that: The perinatal clinical index data include one of the following or a combination thereof: gestational age at birth, birth weight, gender, mode of delivery, APgar1min score, APgar 5min score, etiology, C-reactive protein, and cerebrospinal fluid detection indicators.

Images