Baseline brain CT image processing method and platform and storage medium

Abstract

The invention discloses a baseline brain CT image processing method and platform and a storage medium. The platform comprises an image segmentation module used for obtaining a baseline brain CT image of a user and inputting the baseline brain CT image into a pre-trained segmentation model to obtain a brain parenchyma region bleeding image and a bleeding volume; the image vector construction module is used for inputting the brain parenchyma area bleeding image into a pre-trained radiomics model to obtain a radiomics score; constructing an image vector based on the radiomics score and the bleeding volume; the text vector construction module is used for extracting blood biochemical marker information and medical record information of the user and constructing a text vector; and the output module is used for inputting the image vector and the text vector into a pre-trained risk judgment model and outputting a risk score of the baseline brain CT image. According to the method, the baseline brain CT image can be automatically subjected to quantitative evaluation based on the image processing platform, and the cerebral hemorrhage expansion risk score is output for reference of the user.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and more specifically, to a baseline brain CT image processing method, platform, and storage medium. Background Technology

[0002] Intracerebral hemorrhage is one of the common causes of stroke, and its incidence is increasing year by year with the average age of the population. Increased intracerebral hemorrhage refers to an increase in absolute hemorrhage volume of 6 ml or relative volume of more than 33% compared to baseline during early imaging follow-up after primary intracerebral hemorrhage; this is a predictor of poor prognosis for intracerebral hemorrhage.

[0003] Currently, some hospitals use CTA signs to predict the risk of cerebral hemorrhage expansion, but its application has significant limitations. It requires a head CTA examination, which increases the user's radiation exposure, contrast agent use risks, and financial burden. On the other hand, the definitions of signs on plain CT scans are numerous and overlapping, and their interpretation heavily relies on the radiologist's subjective experience, leading to poor repeatability and unstable predictive efficacy. It also demands a high level of expertise from the evaluating physician, and there is a lack of image processing equipment for automated quantitative evaluation of CT images. Summary of the Invention

[0004] This application provides a baseline brain CT image processing method, platform, and storage medium to at least solve the technical problem in the related art of automatically processing and quantitatively evaluating baseline brain CT images.

[0005] According to one aspect of the embodiments of this application, a baseline brain CT image processing platform is provided, comprising: The image segmentation module is used to acquire the user's baseline brain CT image, input the baseline brain CT image into a pre-trained segmentation model, and obtain the brain parenchymal region hemorrhage image and hemorrhage volume. The image vector construction module is used to input the hemorrhage image of the brain parenchyma region into a pre-trained radiomics model to obtain a radiomics score; and to construct an image vector based on the radiomics score and the hemorrhage volume. The text vector construction module is used to extract users' blood biochemical markers and medical record information to construct text vectors. The output module is used to input the image vector and text vector into a pre-trained risk assessment model and output a risk score of the baseline brain CT image.

[0006] In one implementation, the baseline brain CT image is input into a pre-trained segmentation model to obtain an image of hemorrhage in the brain parenchyma region and the hemorrhage volume, including: The baseline brain CT image is input into a pre-trained segmentation model to segment and obtain an image of the hemorrhage area; Based on the hemorrhage area image, identify the hemorrhage image of the brain parenchyma region; Calculate the hemorrhage volume in the brain parenchyma region.

[0007] In one implementation, the image vector construction module further includes a radiomics model training unit. Used to collect and preprocess a large number of CT images of hemorrhage in brain parenchyma. Feature extraction is performed on the preprocessed image, including but not limited to shape features, texture features, and statistical features; Each image is labeled with a corresponding tag based on medical record information to obtain a training dataset; A machine learning model is trained based on the training dataset to obtain a trained radiomics model.

[0008] In one implementation, the image of the hemorrhage in the brain parenchyma region is input into a pre-trained radiomics model to obtain a radiomics score, including: The images of hemorrhage in the brain parenchyma region are input into a pre-trained radiomics model to automatically quantify and extract image features, including shape features, texture features, and statistical features. The radiomics model uses a built-in list of key features and its pre-trained weights to filter and weight the extracted features. The selected feature values ​​are fused with their corresponding weights to calculate the radiomics score.

[0009] In one implementation, the user's blood biochemical marker information and medical record information are extracted to construct a text vector, including: Extract the user's blood biochemical markers, including coagulation function, complete blood count, and biochemical indicators; Extract the user's medical record information, including onset time, vital signs information, and basic information; The extracted continuous numerical variables are standardized. The standardized numerical features and categorical features are combined to form a feature vector, which is then used to obtain the text vector.

[0010] In one implementation, the image vector and text vector are input into a pre-trained risk assessment model, which outputs a risk score for the baseline brain CT image, including: The image vector and text vector are combined into a single input vector; Load the pre-trained risk assessment model; The input vector is input into the risk assessment model, regression analysis is performed, and the risk score is output.

[0011] In one implementation, the output module further includes a risk assessment model training unit. This is used to collect a large number of baseline brain CT images and medical record information, and to construct image vectors and text vectors based on the CT images and medical record information. Based on the image vectors and text vectors, risk scoring labels are added to obtain the training dataset; A risk assessment model is constructed based on a deep fully connected neural network. The risk assessment model is trained by optimizing the algorithm and adjusting the model parameters to minimize the gap between the predicted risk value and the true label, thus obtaining a well-trained risk assessment model.

[0012] According to another aspect of the embodiments of this application, a baseline brain CT image processing method is provided, comprising: The user's baseline brain CT image is acquired, and the baseline brain CT image is input into a pre-trained segmentation model to obtain images of hemorrhage in the brain parenchyma region and the hemorrhage volume. The brain parenchymal region hemorrhage image is input into a pre-trained radiomics model to obtain a radiomics score; an image vector is constructed based on the radiomics score and the hemorrhage volume; Extract users' blood biochemical markers and medical record information to construct text vectors; The image vectors and text vectors are input into a pre-trained risk assessment model, which outputs a risk score for the baseline brain CT image.

[0013] In one implementation, it further includes: A large number of CT images of hemorrhage in the brain parenchyma were collected and preprocessed. Feature extraction is performed on the preprocessed image, including but not limited to shape features, texture features, and statistical features; Each image is labeled with a corresponding tag based on medical record information to obtain a training dataset; A machine learning model is trained based on the training dataset to obtain a trained radiomics model.

[0014] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, which is configured to execute the above-described baseline brain CT image processing method when running.

[0015] The technical solutions provided in this application embodiment may include the following beneficial effects: This application discloses a baseline brain CT image processing platform that can automatically segment images to obtain images of the hemorrhage area. First, a radiomics model is used to score the hemorrhage area. This radiomics score, as a factor, is input into a risk assessment model along with other clinical information such as blood biochemical markers and medical record information. The platform automatically outputs a risk score characterizing the risk of brain hemorrhage expansion. This application develops an automated and quantitative image processing platform based on plain CT scans, automatically quantifying baseline brain CT images and outputting a brain hemorrhage expansion risk score for user reference. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a baseline brain CT image processing method according to an embodiment of this application; Figure 2 This is a flowchart of another baseline brain CT image processing method according to an embodiment of this application; Figure 3 This is a schematic diagram of a baseline brain CT image processing platform according to an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] This application provides an image processing platform that automatically quantifies and predicts the risk of hemorrhage expansion in a user's CT images by analyzing baseline cerebral hemorrhage CT images, and outputs an image processing score for doctors' reference.

[0020] The baseline brain CT image processing platform of this application embodiment will be described in detail below with reference to the accompanying drawings. Figure 3 As shown, the platform mainly includes the following modules: The image segmentation module 301 is used to acquire the user's baseline brain CT image, input the baseline brain CT image into the pre-trained segmentation model, and obtain the brain parenchymal region hemorrhage image and hemorrhage volume.

[0021] In one implementation, a baseline brain CT image is first input into a pre-trained segmentation model to segment and obtain an image of the hemorrhage region. The segmentation model automatically identifies and segments the image, outputting the hemorrhage region image.

[0022] Furthermore, based on the image of the hemorrhage area, the hemorrhage image of the brain parenchyma region is identified, and the hemorrhage volume of the brain parenchyma region is calculated.

[0023] Specifically, the hemorrhage region image output by the segmentation model is first registered with the original brain CT image, and then spatially filtered using a predefined brain parenchyma mask, such as obtained through brain tissue segmentation or registration with a standard brain atlas, to retain only the hemorrhage region located within the brain parenchyma, thereby obtaining the brain parenchyma hemorrhage image.

[0024] Subsequently, based on parameters such as pixel spacing in the CT images, the number of pixels in the binarized brain parenchymal hemorrhage image is converted into physical spatial volume. By summing the three-dimensional voxel volumes corresponding to the hemorrhage pixels in all consecutive slices, the total hemorrhage volume within the brain parenchyma is calculated. Alternatively, the hemorrhage volume can be calculated using existing CT image processing software; this application does not impose specific limitations on the embodiments described.

[0025] Understandably, the segmentation model is trained before it is loaded.

[0026] Training the segmentation model first requires preparing a dataset of a large number of labeled brain CT images, where the hemorrhage region in each image is precisely delineated as a label. Then, a deep neural network, such as U-Net, is constructed using an encoder-decoder structure. The original brain CT images are used as input, and their corresponding labeled segmentation maps are used as the learning target. During training, the model's predicted segmentation map is obtained through forward propagation, and then a loss function is used to quantify the difference between the predicted map and the true label. The model's weight parameters are iteratively optimized using backpropagation and gradient descent algorithms to minimize this loss function, thereby enabling the model to gradually learn to accurately identify the image features of the hemorrhage region from CT images.

[0027] The image processing platform provided in this application embodiment can automatically segment CT images to obtain the image portion and hemorrhage volume of the brain parenchyma region.

[0028] The image processing platform also includes an image vector construction module 302, which is used to input images of hemorrhage in the brain parenchyma region into a pre-trained radiomics model to obtain a radiomics score; and to construct an image vector based on the radiomics score and the hemorrhage volume.

[0029] In one implementation, the image vector construction module further includes a radiomics model training unit, which is used to collect a large number of CT images of hemorrhage in the brain parenchyma region and preprocess them; extract features based on the preprocessed images, including but not limited to shape features, texture features, and statistical features; add corresponding labels to each image based on medical record information to obtain a training dataset; and train a machine learning model based on the training dataset to obtain a trained radiomics model.

[0030] Specifically, a large number of CT images containing brain parenchymal hemorrhages should be collected. These images can be obtained from medical imaging databases, hospital imaging archives, or publicly available medical imaging resource websites. Ensure that the collected images cover different types of brain parenchymal hemorrhages to increase the diversity and representativeness of the image library, forming an initial image database. Then, preprocess the images.

[0031] Next, a large number of quantitative radiomics features were extracted from the images of hemorrhage in each brain parenchyma region. These features are mainly divided into the following categories: Shape characteristics: Describes the three-dimensional geometric properties of the bleeding area, such as volume, surface area, surface area to volume ratio, compactness, and sphericity. These characteristics directly reflect the size and shape of the bleeding lesion.

[0032] Texture features: describe the spatial distribution and interrelationships of voxel grayscale values ​​within a region of interest, revealing heterogeneous information that is difficult for the human eye to discern. These include the grayscale co-occurrence matrix, grayscale run-length matrix, and grayscale size region matrix.

[0033] Statistical characteristics: Described based on histograms of voxel gray values, such as mean, standard deviation, skewness, kurtosis, etc., reflecting the overall gray value distribution.

[0034] Furthermore, the extracted hundreds or even thousands of feature values ​​are organized to form a feature vector for each sample (i.e., each bleeding area). Simultaneously, a label based on clinical medical record information is added to each sample. For example, it includes information such as whether the bleeding has expanded or whether the prognosis is good. This yields the training dataset.

[0035] Furthermore, the complete training dataset is randomly divided into a training set (typically 70-80%), a validation set (10-15%), and a test set (10-15%). The training set is used for model learning, the validation set is used for hyperparameter tuning and model selection, and the test set is used for final model performance evaluation.

[0036] Further, select a suitable machine learning algorithm. Commonly used algorithms include logistic regression, support vector machines, and random forests. Train the machine learning model based on the training dataset to obtain a trained radiomics model.

[0037] In one implementation, images of hemorrhage in the brain parenchyma region are input into a pre-trained radiomics model to obtain a radiomics score.

[0038] Images of hemorrhage in the brain parenchyma region are input into a radiomics model, which automatically quantifies and extracts image features, including shape, texture, and statistical features. The radiomics model then filters and weights the extracted features based on a built-in list of key features learned during training and its pre-trained weights. The filtered feature values ​​are then fused with their corresponding weights to calculate a radiomics score. This score will serve as a key numerical indicator for downstream prediction tasks.

[0039] Furthermore, image vectors are constructed based on radiomics scores and hemorrhage volume.

[0040] First, the hemorrhage volume is standardized. Then, the standardized volume value and the radiomics score are combined to form a two-dimensional feature vector.

[0041] The image processing platform also includes a text vector construction module 303, which is used to extract the user's blood biochemical marker information and medical record information to construct text vectors.

[0042] In one optional implementation, the user's blood biochemical marker information, including coagulation function, complete blood count, and biochemical indicators, is extracted; the user's medical record information, including onset time, vital signs information, and basic information, such as age and gender, is also extracted.

[0043] In one exemplary scenario, coagulation function includes prothrombin time, activated partial thromboplastin time, international normalized ratio, and fibrinogen; complete blood count includes white blood cell count and platelet count; biochemical indicators include blood glucose, creatinine, and electrolytes. Other parameters may also be included; the embodiments in this application are merely illustrative.

[0044] Key medical record information extraction: For example, time information: time from onset to admission (in hours or minutes). Vital signs: systolic blood pressure, diastolic blood pressure, heart rate, respiratory rate, body temperature, etc. upon admission. Basic information: age, gender.

[0045] Furthermore, the extracted continuous numerical variables are standardized, and the standardized numerical features are combined with the categorical features to form a feature vector, thus obtaining a text vector.

[0046] For example, standardizing all extracted continuous numerical variables (such as blood pressure, blood sugar, and onset time) is commonly done using Z-score standardization or maximum / minimum normalization. This can eliminate the influence of differences in the units and ranges of different indicators.

[0047] Furthermore, the standardized numerical features are combined with necessary categorical features to form a feature vector. This vector can then be directly concatenated to construct a structured numerical vector.

[0048] In another alternative implementation, to construct a text vector containing user blood biochemical marker information and medical record information, relevant text data is first extracted from a medical database. This includes detailed medical record descriptions and specific values ​​and interpretations of blood biochemical markers. This text data is typically in an unstructured form and therefore requires preprocessing, such as removing irrelevant symbols, word segmentation, and stop word removal, to ensure data cleanliness and consistency.

[0049] Next, a pre-trained BERT model is used to process the pre-processed text. The BERT model is a deep learning model based on the Transformer architecture. It is pre-trained on a large amount of text data and is able to understand the semantics and contextual relationships within the text. We input the extracted text data into the BERT model, which outputs a high-dimensional vector representation of each text segment. These vectors capture the complex semantic information in the text, enabling even unstructured text data to be effectively transformed into numerical features usable by machine learning models.

[0050] The image processing platform also includes an output module 304, which is used to input image vectors and text vectors into a pre-trained risk assessment model and output a risk score of the baseline brain CT image.

[0051] In one implementation, image vectors and text vectors are input into a pre-trained risk assessment model, which outputs a risk score for a baseline brain CT image.

[0052] First, the image vector and text vector are merged into a single input vector, which is then loaded into a pre-trained risk assessment model. The input vector is then fed into the risk assessment model for regression analysis, and a risk score is output.

[0053] Specifically, image vectors and text vectors are typically high-dimensional numerical arrays. Using appropriate programming tools (such as Python's NumPy library), these two vectors are concatenated along a specific axis (usually the last axis) to form a combined input vector.

[0054] The pre-trained risk assessment model is loaded from the specified storage path, and the fused input vector obtained in the first step is fed into the pre-trained model loaded in the second step. The model performs a series of calculations on the input vector based on its learned complex nonlinear mapping relationships. The model's output layer produces a continuous real value, i.e., the risk score. This score is the result of a regression analysis, and its numerical range is consistent with the label normalization method used during model training.

[0055] This process effectively integrates multimodal data and leverages the predictive capabilities of pre-trained models to provide accurate risk assessments.

[0056] In one implementation, the output module further includes a risk assessment model training unit, which collects a large number of baseline brain CT images and medical record information, and constructs image vectors and text vectors based on the CT images and medical record information.

[0057] Furthermore, risk scoring labels are added based on image vectors and text vectors to obtain the training dataset; Furthermore, a suitable model structure for the regression task is selected. In one implementation, a deep fully connected neural network is used, including an input layer (the number of neurons must be equal to the dimension of the fused vector), hidden layers (containing several fully connected layers, each followed by an activation function (such as ReLU) to introduce non-linearity), and an output layer (one neuron using a linear activation function to directly output a continuous predicted risk value). This yields the constructed risk assessment model.

[0058] Finally, the risk assessment model was trained by optimizing the algorithm and adjusting the model parameters to minimize the gap between the predicted risk value and the true label, thus obtaining a well-trained risk assessment model.

[0059] The loss function measures the difference between the model's predicted value and the true label. For regression tasks, mean squared error can be used as the loss function. The model is then trained using optimization algorithms, such as gradient descent, to adjust the model parameters and minimize the difference between the predicted risk value and the true label. Finally, after sufficient training and parameter tuning, a performance-optimized risk assessment model is obtained, which can be used to predict risk scores for new input data.

[0060] The output module fully leverages the rich features of medical images and medical records by integrating image and text vectors. This multimodal data fusion method can more comprehensively reflect the risk of bleeding expansion in images, improving the accuracy and reliability of risk assessment. A well-trained risk assessment model can quickly and accurately output risk scores, providing crucial decision support for clinicians.

[0061] The image processing platform of this application first analyzes brain CT image data using a radiomics model to predict a radiomics score. This score, as an important predictive factor, is incorporated into a comprehensive predictive model along with other clinical risk factors (such as hemorrhage volume and onset time). By integrating this multi-dimensional information, the model can more comprehensively assess the risk of hemorrhage expansion and ultimately output an accurate risk value. This risk value is provided to doctors for reference, who then use it to further determine whether the patient's brain hemorrhage has expanded. This application only protects the image processing platform itself.

[0062] According to another aspect of the embodiments of this application, a baseline brain CT image processing method for implementing the above-described baseline brain CT image processing platform is also provided. For example... Figure 1 As shown, the method includes: S101 acquires the user's baseline brain CT image, inputs the baseline brain CT image into a pre-trained segmentation model, and obtains images of hemorrhage in the brain parenchyma region and the hemorrhage volume. S102 inputs images of hemorrhage in the brain parenchyma region into a pre-trained radiomics model to obtain a radiomics score; and constructs an image vector based on the radiomics score and the hemorrhage volume. S103 extracts the user's blood biochemical markers and medical record information to construct a text vector; S104 inputs image vectors and text vectors into a pre-trained risk assessment model and outputs a risk score for the baseline brain CT image.

[0063] In one implementation, the method further includes: collecting a large number of CT images of hemorrhage in the brain parenchyma and preprocessing them; extracting features based on the preprocessed images, including but not limited to shape features, texture features, and statistical features; adding corresponding labels to each image based on medical record information to obtain a training dataset; and training a machine learning model based on the training dataset to obtain a trained radiomics model.

[0064] To facilitate understanding of the methods in the embodiments of this application, the following description is provided in conjunction with the appendix. Figure 2 Further description. For example... Figure 2 As shown, the steps include: S201 acquires the user's baseline brain CT image, inputs the baseline brain CT image into a pre-trained segmentation model, and obtains images of hemorrhage in the brain parenchyma region and the hemorrhage volume. S202 is based on a machine learning model and trains a radiomics model. S203 inputs images of hemorrhage in the brain parenchyma region into a pre-trained radiomics model to obtain a radiomics score; S204 constructs image vectors based on radiomics scoring and hemorrhage volume; S205 extracts users' blood biochemical markers and medical record information to construct text vectors; S206 inputs image vectors and text vectors into a pre-trained risk assessment model, performs regression analysis, and outputs a risk score.

[0065] The image processing platform provided in this application achieves a comprehensive and accurate assessment of the prognostic risk of cerebral hemorrhage by deeply integrating radiomics and clinical data. The platform first uses a pre-trained segmentation model to automatically and accurately delineate the hemorrhage area. Then, it extracts deep quantitative features from the hemorrhage area—features difficult for the human eye to discern—using a radiomics model to generate an objective radiomics score, effectively avoiding the over-reliance on physician experience in traditional image interpretation. By multimodal fusion of radiomics scores, hemorrhage volume, and other image features with structured clinical text vectors, the platform constructs a comprehensive predictive model that surpasses single data sources. Ultimately, the platform can quickly and automatically output a quantitative risk score, providing clinicians with repeatable and highly accurate decision support, and has significant clinical application value.

[0066] It should be noted that the baseline brain CT image processing platform provided in the above embodiments is only illustrated by the division of the above functional modules when executing the baseline brain CT image processing method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the baseline brain CT image processing platform and the baseline brain CT image processing method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the platform embodiments, which will not be repeated here.

[0067] According to another aspect of the present application, a computer-readable storage medium corresponding to the baseline brain CT image processing method provided in the foregoing embodiments is also provided, wherein a computer program (i.e., a program product) is stored thereon, and when the computer program is run by a processor, it executes the baseline brain CT image processing method provided in any of the foregoing embodiments.

[0068] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0069] The computer-readable storage medium provided in the above embodiments of this application and the baseline brain CT image processing method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A baseline brain CT image processing platform, characterized in that, include: The image segmentation module is used to acquire the user's baseline brain CT image, input the baseline brain CT image into a pre-trained segmentation model, and obtain the brain parenchymal region hemorrhage image and hemorrhage volume. The image vector construction module is used to input the hemorrhage image of the brain parenchyma region into a pre-trained radiomics model to obtain a radiomics score; and to construct an image vector based on the radiomics score and the hemorrhage volume. The text vector construction module is used to extract users' blood biochemical markers and medical record information to construct text vectors. The output module is used to input the image vector and text vector into a pre-trained risk assessment model and output a risk score of the baseline brain CT image.

2. The image processing platform according to claim 1, characterized in that, The baseline brain CT image is input into a pre-trained segmentation model to obtain images of hemorrhage in the brain parenchyma region and the hemorrhage volume, including: The baseline brain CT image is input into a pre-trained segmentation model to segment and obtain an image of the hemorrhage area; Based on the hemorrhage area image, identify the hemorrhage image of the brain parenchyma region; Calculate the hemorrhage volume in the brain parenchyma region.

3. The image processing platform according to claim 1, characterized in that, The image vector construction module also includes a radiomics model training unit. Used to collect and preprocess a large number of CT images of hemorrhage in brain parenchyma. Feature extraction is performed on the preprocessed image, including but not limited to shape features, texture features, and statistical features; Each image is labeled with a corresponding tag based on medical record information to obtain a training dataset; A machine learning model is trained based on the training dataset to obtain a trained radiomics model.

4. The image processing platform according to claim 1, characterized in that, The images of hemorrhage in the brain parenchyma region are input into a pre-trained radiomics model to obtain a radiomics score, including: The images of hemorrhage in the brain parenchyma region are input into a pre-trained radiomics model to automatically quantify and extract image features, including shape features, texture features, and statistical features. The radiomics model uses a built-in list of key features and its pre-trained weights to filter and weight the extracted features. The selected feature values ​​are fused with their corresponding weights to calculate the radiomics score.

5. The image processing platform according to claim 1, characterized in that, Extract users' blood biochemical markers and medical record information, and construct text vectors, including: Extract the user's blood biochemical markers, including coagulation function, complete blood count, and biochemical indicators; Extract the user's medical record information, including onset time, vital signs information, and basic information; The extracted continuous numerical variables are standardized. The standardized numerical features and categorical features are combined to form a feature vector, which is then used to obtain the text vector.

6. The image processing platform according to claim 1, characterized in that, The image vectors and text vectors are input into a pre-trained risk assessment model, which outputs a risk score for the baseline brain CT image, including: The image vector and text vector are combined into a single input vector; Load the pre-trained risk assessment model; The input vector is input into the risk assessment model, regression analysis is performed, and the risk score is output.

7. The image processing platform according to claim 1, characterized in that, The output module also includes a risk assessment model training unit. This is used to collect a large number of baseline brain CT images and medical record information, and to construct image vectors and text vectors based on the CT images and medical record information. Based on the image vectors and text vectors, risk scoring labels are added to obtain the training dataset; A risk assessment model is constructed based on a deep fully connected neural network. The risk assessment model is trained by optimizing the algorithm and adjusting the model parameters to minimize the gap between the predicted risk value and the true label, thus obtaining a well-trained risk assessment model.

8. A baseline brain CT image processing method, characterized in that, include: The user's baseline brain CT image is acquired, and the baseline brain CT image is input into a pre-trained segmentation model to obtain images of hemorrhage in the brain parenchyma region and the hemorrhage volume. The brain parenchymal region hemorrhage image is input into a pre-trained radiomics model to obtain a radiomics score; an image vector is constructed based on the radiomics score and the hemorrhage volume; Extract users' blood biochemical markers and medical record information to construct text vectors; The image vectors and text vectors are input into a pre-trained risk assessment model, which outputs a risk score for the baseline brain CT image.

9. The method according to claim 8, characterized in that, Also includes: A large number of CT images of hemorrhage in the brain parenchyma were collected and preprocessed. Feature extraction is performed on the preprocessed image, including but not limited to shape features, texture features, and statistical features; Each image is labeled with a corresponding tag based on medical record information to obtain a training dataset; A machine learning model is trained based on the training dataset to obtain a trained radiomics model.

10. A computer-readable medium, characterized in that, It stores computer-readable instructions that are executed by a processor to implement a baseline brain CT image processing method as described in any one of claims 8 to 9.

Images