Deep learning-based intracranial hemorrhage prognosis prediction method and system

Abstract

The invention discloses an intracranial hemorrhage prognosis prediction method and system based on deep learning, and the method comprises the steps: carrying out the preprocessing of a brain CT image, so as to obtain first preprocessing data, and carrying out the preprocessing of clinical data, so as to obtain second preprocessing data; constructing a fusion model, and extracting a feature vector of the first preprocessed data through a convolutional neural network to obtain a first feature vector; extracting a feature vector of the second pre-processed data through a feedforward neural network to obtain a second feature vector; splicing the first feature vector and the second feature vector to obtain a comprehensive feature vector; and performing data preprocessing on a to-be-predicted brain CT image and clinical data corresponding to the to-be-predicted brain CT image, and inputting the preprocessed data into the optimal fusion model to obtain a prognosis prediction result. By applying the technical scheme, the problem that in the prior art, the intracranial hemorrhage prognosis prediction accuracy is poor is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting the prognosis of intracranial hemorrhage based on deep learning. Background Art

[0002] At present, the application of deep learning technology in the field of medical imaging is becoming more and more extensive. In the analysis of head CT scan images, deep learning models have been successfully used to detect and identify various lesions, including intracranial hemorrhage. The existing models can automatically identify and locate abnormal areas by learning a large amount of medical imaging data, providing doctors with important auxiliary information. However, the existing models mainly focus on the detection of intracranial hemorrhage, and there are relatively few studies on prognosis prediction, which leads to the lack of accurate prognostic information for clinicians when facing patients with intracranial hemorrhage, affecting the making of clinical decisions.

[0003] In addition, existing technologies often process medical images and clinical data separately, lack comprehensive analysis of the relationship between the two, and have limitations in processing the combined input of medical images and clinical data. Due to the high cost of acquiring and annotating medical image data, data sets are often limited in size and difficult to meet the training needs of deep learning models.

[0004] Therefore, there is an urgent need for a method for predicting the prognosis of intracranial hemorrhage that can integrate brain CT images and the clinical data and utilize deep learning technology to improve the accuracy of intracranial hemorrhage prognosis prediction. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and system for predicting the prognosis of intracranial hemorrhage based on deep learning, so as to solve the problem of poor accuracy of prognosis prediction of intracranial hemorrhage in the prior art.

[0006] To achieve the above object, an embodiment of the present invention provides a method for predicting the prognosis of intracranial hemorrhage based on deep learning, and the method specifically includes: Acquiring a brain CT image and clinical data corresponding to the brain CT image, preprocessing the brain CT image to obtain first preprocessed data, and preprocessing the clinical data to obtain second preprocessed data; Constructing a fusion model, wherein the fusion model includes a convolutional neural network and a feedforward neural network; Extracting a feature vector of the first preprocessed data by a convolutional neural network to obtain a first feature vector; Extracting a feature vector of the second preprocessed data by a feedforward neural network to obtain a second feature vector; Concatenate the first feature vector and the second feature vector to obtain a comprehensive feature vector; Using the comprehensive feature vector as an input of a fusion model, and training the fusion model to obtain an optimal fusion model; The brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted are preprocessed and then input into the optimal fusion model to obtain a prognosis prediction result.

[0007] On the basis of the above technical solution, the present invention can also be improved as follows: Optionally, preprocessing the brain CT image to obtain first preprocessed data includes: The brain CT image is subjected to size adjustment, pixel value normalization, background noise removal and contrast enhancement.

[0008] Optionally, the contrast enhancement processing includes: Divide the input image into several non-overlapping sub-blocks, calculate the grayscale histogram of each sub-block, and set a cropping limit , the number of pixels in the histogram greater than the clipping limit The gray levels are cropped and the cropped pixels are redistributed to other gray levels to limit the contrast of the brain CT image: The cropped histogram is calculated using formula (1): Formula (1); In the formula, is the cropped histogram, For the The histogram of the sub-blocks, The histogram The number of pixels in grayscale level, is the clipping limit, is the number of gray levels, k is the index of each gray level in the histogram; The grayscale value after equalization is calculated by formula (2): Formula (2); In the formula, is the gray value after equalization, Indicates rounding down; The gray value of the pixel after enhancement is calculated by formula (3); Formula (3); In the formula, is a pixel, is the gray value of the pixel after enhancement, is the interpolation weight, is the equalization mapping value of the four sub-blocks where the pixel is located.

[0009] Optionally, preprocessing the clinical data to obtain second preprocessed data includes: The standard scores of the variables were calculated by formula (4); Formula (4); Where Z is the standard score of the variable, is the original variable value, is the mean of the variable, is the standard deviation of the variable; The clinical data are converted into standard scores by formula (4), so that different clinical data variables can be compared and processed on the same scale.

[0010] Optionally, extracting the feature vector of the first preprocessed data by a convolutional neural network to obtain a first feature vector includes: The output of the convolutional layer is calculated using formula (5); Formula (5); In the formula, is the output of the convolutional layer, is the convolution kernel, is the bias and ReLU is the activation function.

[0011] Optionally, extracting the feature vector of the second preprocessed data by a feedforward neural network to obtain the second feature vector includes: The output of the hidden layer is calculated by formula (6); Formula (6); In the formula, is the output of the hidden layer, is the input of the hidden layer, represents the weight matrix, Represents the bias vector.

[0012] Optionally, the brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted are preprocessed, inputted into the optimal fusion model, and the prediction result is outputted, including: The prognostic probability was calculated by formula (7); Formula (7); In the formula, is the prognostic probability, is the Sigmoid function, is the weight matrix, is the bias vector, and H is the comprehensive feature vector.

[0013] A system for predicting the prognosis of intracranial hemorrhage based on deep learning, comprising: An acquisition module, used for acquiring a brain CT image and clinical data corresponding to the brain CT image; A data processing module, configured to preprocess the brain CT image to obtain first preprocessed data, and preprocess the clinical data to obtain second preprocessed data; A construction module, used to construct a fusion model, wherein the fusion model includes a convolutional neural network and a feedforward neural network; Extracting a feature vector of the first preprocessed data by a convolutional neural network to obtain a first feature vector; Extracting a feature vector of the second preprocessed data by a feedforward neural network to obtain a second feature vector; A feature concatenation module, used for concatenating the first feature vector and the second feature vector to obtain a comprehensive feature vector; A training module, used for taking the comprehensive feature vector as an input of a fusion model and training the fusion model to obtain an optimal fusion model; The prediction module is used to pre-process the brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted, and then input them into the optimal fusion model to obtain the prognosis prediction result.

[0014] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0015] A non-transitory computer-readable storage medium stores a computer program, which implements the steps of the method when executed by a processor.

[0016] The embodiments of the present invention have the following advantages: The method for predicting the prognosis of intracranial hemorrhage based on deep learning in the present invention combines brain CT images and mixed brain CT images into a deep learning model, so that the model can comprehensively evaluate the patient's condition from multiple angles, improve the accuracy and reliability of prognosis prediction, not only improve the performance of the model, but also simplify the decision-making process of doctors, help improve clinical work efficiency, and the mixed input structure makes data utilization more efficient, saving the time cost of data processing and analysis; The present invention uses a large amount of medical imaging data for training, and automatically learns features and patterns through a deep learning network, without manual intervention for feature selection, thereby improving the generalization ability and prediction accuracy of the model. Compared with traditional manual feature extraction methods, deep learning models can better capture complex associations in data and improve the accuracy and efficiency of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the implementation mode of the present invention or the technical solution in the prior art, the following will briefly introduce the drawings required for the implementation mode or the prior art description. Obviously, the drawings described below are only exemplary. For ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work. The structure, proportion, size, etc. illustrated in this specification are only used to match the content disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can be implemented, so it has no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effect that the present invention can produce and the purpose that can be achieved, should still fall within the scope of the technical content disclosed in the present invention.

[0018] Figure 1 Schematic diagram of a process for predicting the prognosis of intracranial hemorrhage based on deep learning in an embodiment of the present invention; Figure 2 A schematic diagram of the main components of a system for predicting the prognosis of intracranial hemorrhage based on deep learning in an embodiment of the invention; Figure 3 It is a structural diagram of the fusion model of the present invention; Figure 4 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Figure 1 FIG. 1 is a flow chart of a method for predicting the prognosis of intracranial hemorrhage based on deep learning in an embodiment of the present invention. Figure 1 As shown, a method for predicting the prognosis of intracranial hemorrhage based on deep learning provided by an embodiment of the present invention includes the following steps S101 to S107.

[0021] S101, acquiring a brain CT image and clinical data corresponding to the brain CT image, preprocessing the brain CT image to obtain first preprocessed data, and preprocessing the clinical data to obtain second preprocessed data.

[0022] The above-mentioned preprocessing may specifically include: resizing the brain CT image, normalizing pixel values, removing background noise, and enhancing contrast.

[0023] The brain CT images were resized and pixel values ​​were normalized to ensure consistency of the model input. Specifically, the image size was uniformly adjusted to 128×128×128 pixels, and the pixel values ​​were normalized to the range of [0, 1] to reduce noise and imbalance in the data.

[0024] By removing background noise and enhancing contrast, the visibility of the bleeding area is enhanced, thereby improving the model's ability to extract features.

[0025] The contrast enhancement process comprises: Contrast Limited Adaptive Histogram Equalization (CLAHE) is used. The CLAHE method first divides the input image into several non-overlapping tiles, each of which is . Calculate the grayscale histogram of each sub-block, assuming The histogram of the sub-block is , its The number of pixels in gray level is The histogram of each sub-block is clipped to limit the contrast. Specifically, a clipping limit is set , the number of pixels in the histogram greater than the clipping limit The gray levels are cropped and the cropped pixels are redistributed to other gray levels to limit the contrast of the brain CT image: The cropped histogram is calculated by formula (1) to limit the contrast of the image: Formula (1); In the formula, is the cropped histogram, For the The histogram of the sub-blocks, The histogram The number of pixels in grayscale level, is the clipping limit, is the number of gray levels, k is the index of each gray level in the histogram; The grayscale value after equalization is calculated by formula (2) and used to implement histogram equalization to enhance the contrast of the image: Formula (2); In the formula, is the gray value after equalization, Indicates rounding down; For each pixel, its enhanced grayscale value is calculated by bilinear interpolation according to the equalization mapping value of the sub-block to which it belongs.

[0026] By using formula (3) to calculate the grayscale value of the enhanced pixel, the boundary artifacts caused by sub-block division can be effectively reduced, making the enhanced image visually smoother and more natural, and better able to present the details in the image, especially in brain CT images, which helps to more accurately observe the subtle structures and lesions of the brain of ICH patients; Formula (3); In the formula, is a pixel, is the gray value of the pixel after enhancement, is the interpolation weight, is the equalization mapping value of the four sub-blocks where the pixel is located.

[0027] Clinical data related to ICH patients, including age, gender, blood pressure, blood glucose level, and past medical history, were z-score standardized.

[0028] The standard scores of the variables were calculated by formula (4); Formula (4); Where Z is the standard score of the variable, is the original variable value, is the mean of the variable, This standardization method makes the mean of each variable 0 and the standard deviation 1, thus eliminating the dimensional differences between different variables.

[0029] The clinical data are converted into standard scores by formula (4), so that different clinical data variables can be compared and processed on the same scale.

[0030] S102, constructing a fusion model, wherein the fusion model includes a convolutional neural network and a feedforward neural network.

[0031] S103, extracting a feature vector of the first preprocessed data through a convolutional neural network to obtain a first feature vector.

[0032] Specifically, Figure 3 As shown in Figure 1, multiple 3D convolutional layers are used to extract the spatial features of brain CT images. The output of each convolutional layer is nonlinearly transformed through the ReLU activation function, and the input image size is 128×128×128; The output of the convolutional layer is calculated using formula (5); Formula (5); In the formula, is the output of the convolutional layer, is the convolution kernel, is the bias and ReLU is the activation function.

[0033] Subsequently, downsampling is performed through the maximum pooling layer to reduce the size of the feature map and improve computational efficiency. After flattening the feature map output by the convolutional layer, it is input into the fully connected layer and the feature vector is output through the ReLU activation function. The formula of the fully connected layer is as follows: ; in, is the output of the fully connected layer, represents the weight matrix of the fully connected layer, represents the bias vector, is the feature after flattening operation, is the activation function.

[0034] S104, extracting a feature vector of the second preprocessed data through a feedforward neural network to obtain a second feature vector; Specifically, Figure 3 As shown in the figure, the input layer is the first fully connected layer of the network, which accepts the standardized clinical variables. It is followed by a series of fully connected layers as hidden layers, and each layer uses the ReLU activation function.

[0035] The output of the hidden layer is calculated by formula (6); Formula (6); In the formula, is the output of the hidden layer, is the input of the hidden layer, represents the weight matrix, Represents the bias vector.

[0036] S105, concatenating the first feature vector and the second feature vector to obtain a comprehensive feature vector; Specifically, the feature vector output by CNN and the feature vector output by the feedforward neural network are concatenated to form a mixed input: ; in, is the output of the convolutional layer, is the output of the hidden layer, [ ] represents the concatenation operation of vectors. and Concatenate to obtain a new comprehensive vector H. This concatenation operation is usually used to fuse feature information from different sources. For example, in multimodal data processing, image features (obtained through convolutional layers) and non-image features (obtained through hidden layers) are combined to form a more comprehensive feature representation so that subsequent models (such as classifiers, predictors, etc.) can use these comprehensive features for more accurate analysis and prediction. For example, in the scenario of intracranial hemorrhage prognosis prediction, the H vector may be input into a fully connected layer for the final prognosis prediction.

[0037] S106, using the comprehensive feature vector as the input of the fusion model, and training the fusion model to obtain an optimal fusion model; Specifically, the comprehensive feature vector is divided into a training set, a validation set, and a test set; Train the fusion model based on the training set; Perform performance evaluation on the trained fusion model based on the validation set to obtain a fusion model that meets the performance requirements; The prediction results of the fusion model that meets the performance conditions are evaluated based on the test set, and the evaluation index corresponding to the fusion model is obtained.

[0038] The prediction error is evaluated using the cross entropy loss function, which is used to measure the gap between the model prediction value and the true label. The formula is as follows: ; in, is the value of cross entropy loss, which is an indicator used to quantify the difference between the model prediction result and the true label. The smaller the loss value, the more accurate the model prediction; the larger the loss value, the greater the difference between the model prediction result and the true situation. Indicates the number of samples. When calculating the cross entropy loss, it is usually calculated for a batch of samples, which is the total number of samples in this batch. For example, when training a deep learning model, 64 samples may be input each time for a parameter update. is the true label of the i-th sample, is the predicted probability of the model for the i-th sample. The cross entropy loss function penalizes samples with large prediction errors and guides the model to continuously adjust parameters to improve prediction accuracy.

[0039] The Adam optimizer is used to adjust the model parameters to minimize the loss function. The Adam optimizer combines the advantages of momentum and adaptive learning rate and updates the weights through the following formula: ; in, represents the model parameters, represents the learning rate, and denote the first-order and second-order momentum estimates, respectively, Is a small constant used to prevent the denominator from being zero. In each iteration, the gradient of the loss function with respect to the model parameters is first calculated: ; The gradient is calculated by the back-propagation algorithm to ensure that the gradient calculation is effective and accurate. The parameters are updated using the first-order and second-order momentum estimates in the Adam optimizer. The specific steps are as follows: ; ; in, It is a first-order momentum estimate, which is used to accumulate information about previous gradients in the optimization algorithm. It helps to accelerate the convergence of the model in flat areas and reduce gradient oscillations. It is a second-order momentum estimate, which is used to measure the accumulation of the square of the gradient. It can adaptively adjust the learning rate, reduce the step size in the direction of the larger gradient, and increase the step size in the direction of the smaller gradient. and are the decay rates of the first-order and second-order momentum estimates, usually 0.9 and 0.999. In this way, the model parameters can be adjusted in the direction of reducing the loss function in each iteration, thereby gradually improving the prediction performance of the model.

[0040] The stability and generalization ability of the model were evaluated using 10-fold cross validation. The model performance was evaluated by indicators such as AUC and accuracy. The model proposed in the present invention had an AUC of 0.924 and an accuracy of 0.861 on the validation set, which was significantly better than the model using only images or only clinical data. The model not only improves the prediction accuracy, but also provides an important reference for clinical decision-making, and has broad application prospects.

[0041] S107, preprocessing the brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted, inputting the data into the optimal fusion model, and outputting the prognosis prediction result; Specifically, Figure 3 As shown, the mixed input passes through the last fully connected layer (output layer) and uses the Sigmoid activation function to output the prognosis probability: The prognostic probability is calculated by formula (7): Formula (7); In the formula, is the prognostic probability, is the Sigmoid function, is the weight matrix, is the bias vector, and H is the comprehensive feature vector. The Sigmoid function limits the output probability to the range of [0, 1] and is used to represent the patient's prognosis probability.

[0042] Figure 2 Flow chart of a system embodiment of the present invention for predicting the prognosis of intracranial hemorrhage based on deep learning; Figure 2 As shown, a system 1 for predicting the prognosis of intracranial hemorrhage based on deep learning provided by an embodiment of the present invention includes an acquisition module 10, a data processing module 20, a construction module 30, a feature splicing module 40, a training module 50 and a prediction module 60: The acquisition module 10 is used to acquire a certain number of brain CT images and clinical data corresponding to the brain CT images.

[0043] The data processing module 20 is used to preprocess the above-mentioned brain CT image to obtain first preprocessed data, and to preprocess the above-mentioned clinical data to obtain second preprocessed data.

[0044] The data processing module 20 is also used to resize the brain CT image, normalize the pixel values, remove background noise, and enhance contrast. The contrast enhancement process includes: dividing the input image into a number of non-overlapping sub-blocks, calculating the grayscale histogram of each sub-block, setting a clipping limit , the number of pixels in the histogram greater than the clipping limit The gray level is cropped and the cropped pixels are redistributed to other gray levels to limit the contrast of the brain CT image: the cropped histogram is calculated by formula (1), the equalized gray value is calculated by formula (2), the enhanced gray value of the pixel is calculated by formula (3), and the standard score of the variable is calculated by formula (4).

[0045] The construction module 30 is used to construct a fusion model, which includes a convolutional neural network and a feedforward neural network; extracting the feature vector of the first preprocessed data through the convolutional neural network to obtain a first feature vector; calculating the output of the convolution layer through formula (5); extracting the feature vector of the second preprocessed data through the feedforward neural network to obtain a second feature vector.

[0046] The feature concatenation module 40 is used to concatenate the first feature vector and the second feature vector to obtain a comprehensive feature vector.

[0047] The training module 50 is used to use the above comprehensive feature vector as the input of the fusion model and train the fusion model to obtain the optimal fusion model; The prediction module 60 is used to input the to-be-predicted brain CT image and the clinical data corresponding to the to-be-predicted brain CT image into the optimal fusion model after data preprocessing to obtain the prognosis prediction result.

[0048] The prediction module 60 is also used to calculate the prognosis probability using formula (7).

[0049] Figure 4 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention, such as Figure 4 As shown, the electronic device 70 includes: a processor 701 (processor), a memory 702 (memory) and a bus 703; The processor 701 and the memory 702 communicate with each other via the bus 703; The processor 701 is used to call the program instructions in the memory 702 to execute the method provided by the embodiment of the present invention.

[0050] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the method provided by the embodiment of the present invention.

[0051] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various storage media that can store program codes.

[0052] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0053] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0054] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for predicting the prognosis of intracranial hemorrhage based on deep learning, characterized in that: include: Acquiring a brain CT image and clinical data corresponding to the brain CT image, preprocessing the brain CT image to obtain first preprocessed data, and preprocessing the clinical data to obtain second preprocessed data; Constructing a fusion model, wherein the fusion model includes a convolutional neural network and a feedforward neural network; Extracting a feature vector of the first preprocessed data by a convolutional neural network to obtain a first feature vector; Extracting a feature vector of the second preprocessed data by a feedforward neural network to obtain a second feature vector; Concatenate the first feature vector and the second feature vector to obtain a comprehensive feature vector; Using the comprehensive feature vector as an input of a fusion model, and training the fusion model to obtain an optimal fusion model; The brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted are preprocessed and then input into the optimal fusion model to obtain a prognosis prediction result.

2. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 1, characterized in that: The preprocessing of the brain CT image to obtain first preprocessed data includes: The brain CT image is subjected to size adjustment, pixel value normalization, background noise removal and contrast enhancement.

3. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 2, characterized in that: The contrast enhancement process includes: Divide the input image into several non-overlapping sub-blocks, calculate the grayscale histogram of each sub-block, and set a cropping limit , the number of pixels in the histogram greater than the clipping limit The gray levels are cropped and the cropped pixels are redistributed to other gray levels to limit the contrast of the brain CT image: The cropped histogram is calculated using formula (1): Formula (1): In the formula, is the cropped histogram, For the The histogram of the sub-blocks, The histogram The number of pixels in grayscale level, is the clipping limit, is the number of gray levels, k is the index of each gray level in the histogram; The grayscale value after equalization is calculated by formula (2): Formula (2): In the formula, is the gray value after equalization, Indicates rounding down; The gray value of the pixel after enhancement is calculated by formula (3); Formula (3): In the formula, is a pixel, is the gray value of the pixel after enhancement, is the interpolation weight, is the equalization mapping value of the four sub-blocks where the pixel is located.

4. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 1, characterized in that: The preprocessing of the clinical data to obtain second preprocessed data includes: The standard scores of the variables were calculated by formula (4); Formula (4): Where Z is the standard score of the variable, is the original variable value, is the mean of the variable, is the standard deviation of the variable; The clinical data are converted into standard scores by formula (4), so that different clinical data variables can be compared and processed on the same scale.

5. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 1, characterized in that: The step of extracting the feature vector of the first preprocessed data by using a convolutional neural network to obtain a first feature vector includes: The output of the convolutional layer is calculated using formula (5); Formula (5): In the formula, is the output of the convolutional layer, is the convolution kernel, is the bias and ReLU is the activation function.

6. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 5, characterized in that: The step of extracting the feature vector of the second preprocessed data by a feedforward neural network to obtain a second feature vector includes: The output of the hidden layer is calculated by formula (6); Formula (6): In the formula, is the output of the hidden layer, is the input of the hidden layer, represents the weight matrix, Represents the bias vector.

7. The method for predicting the prognosis of intracranial hemorrhage based on deep learning according to claim 1, characterized in that: The method preprocesses the brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted, inputs them into the optimal fusion model, and outputs the prediction result, including: The prognostic probability is calculated by formula (7): Formula (7): In the formula, is the prognostic probability, is the Sigmoid function, is the weight matrix, is the bias vector, and H is the comprehensive feature vector.

8. A system for predicting the prognosis of intracranial hemorrhage based on deep learning, characterized in that: include: An acquisition module, used for acquiring a brain CT image and clinical data corresponding to the brain CT image; A data processing module, configured to preprocess the brain CT image to obtain first preprocessed data, and preprocess the clinical data to obtain second preprocessed data; A construction module, used to construct a fusion model, wherein the fusion model includes a convolutional neural network and a feedforward neural network; Extracting a feature vector of the first preprocessed data by a convolutional neural network to obtain a first feature vector; Extracting a feature vector of the second preprocessed data by a feedforward neural network to obtain a second feature vector; A feature concatenation module, used for concatenating the first feature vector and the second feature vector to obtain a comprehensive feature vector; A training module, used for taking the comprehensive feature vector as an input of a fusion model and training the fusion model to obtain an optimal fusion model; The prediction module is used to pre-process the brain CT image to be predicted and the clinical data corresponding to the brain CT image to be predicted, and then input them into the optimal fusion model to obtain the prognosis prediction result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images