Method and device for predicting severe pneumonia, electronic equipment and medium

Abstract

The present application relates to the technical field of CT image analysis and processing, and discloses a method and device for predicting the severity of pneumonia, electronic equipment and a medium. The method comprises: collecting a three-dimensional lung CT image of a patient, pre-processing the three-dimensional lung CT image, and obtaining a corresponding two-dimensional CT lung region image; based on the two-dimensional CT lung region image, performing hierarchical prediction through a prediction model, and outputting a first prediction result of each CT slice in the two-dimensional CT lung region image; based on the first prediction result of each CT slice, performing severity prediction through a preset logistic regression model, and outputting a corresponding severity weight; predicting the three-dimensional lung CT image, obtaining a corresponding pneumonia evaluation value, correcting the pneumonia evaluation value and the severity weight, and determining a final prediction result; thereby improving the accuracy of the severity prediction of pneumonia, and improving the performance and medical interpretability of the machine learning model.

Description

Technical Field

[0001] This invention relates to the field of CT image analysis and processing technology, and in particular to a method, device, electronic device, and medium for predicting the severity of pneumonia. Background Technology

[0002] Pneumonia is a lung infection caused by pathogens such as bacteria and viruses. Because these pathogens spread easily among people, a large number of patients can appear in a short period, overwhelming the healthcare system. Therefore, a rapid method for predicting the severity of pneumonia is urgently needed for the rapid triage of patients.

[0003] Currently, traditional medical monitoring indicators require infectious disease expert groups to conduct extensive testing and screening and hold related discussions in advance. However, machine learning technology based solely on medical images cannot be widely applied due to its unstable performance and lack of interpretability. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a pneumonia severity prediction device, electronic device and medium that overcomes or at least partially solves the above problems.

[0005] To achieve the above objectives, the present invention provides a method for predicting the severity of pneumonia, the method comprising:

[0006] Acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images;

[0007] Based on the two-dimensional CT lung region image, a prediction model is used to perform layered prediction, and the first prediction result of each CT slice in the two-dimensional CT lung region image is output.

[0008] Based on the first prediction result of each CT slice, a pre-set logistic regression model is used to predict the severity of the disease and output the corresponding severity weight.

[0009] The three-dimensional lung CT images are used to predict the corresponding pneumonia assessment value, and the pneumonia assessment value and the severity weight are corrected to determine the final prediction result.

[0010] Preferably, before the step of acquiring the patient's three-dimensional lung CT images and preprocessing the three-dimensional lung CT images to obtain the corresponding three-dimensional CT compressed images, the method further includes:

[0011] Multiple different sample two-dimensional CT lung region images are acquired, and the sample two-dimensional CT lung region images are used to construct a training set for training a preset number of different two-dimensional convolutional neural network models;

[0012] Obtain the sample annotation results corresponding to the two-dimensional CT lung region images of each sample;

[0013] The sample 2D CT lung region images in the training set are used as inputs to a predetermined number of different 2D convolutional neural network models. The sample annotation results are used as outputs to the predetermined number of initial models. The predetermined number of different 2D convolutional neural network models are iteratively trained until the accuracy of the validation set of the 2D convolutional neural network model reaches a stable level after iterative training, at which point training stops and the corresponding prediction model is obtained.

[0014] Preferably, before the step of acquiring the patient's three-dimensional lung CT images and preprocessing the three-dimensional lung CT images to obtain the corresponding two-dimensional CT lung region images, the method further includes:

[0015] Multiple different sample 2D CT lung region images are acquired, and the prediction model is used to perform layered prediction on the multiple different sample 2D CT lung region images to obtain the sample prediction results of different layer slices corresponding to each sample 2D CT lung region image.

[0016] Based on the sample prediction results of the different slices, the initial logistic regression model is trained and the severity of the illness is regressed to obtain the trained logistic regression model.

[0017] Preferably, the step of preprocessing the three-dimensional lung CT image to obtain the corresponding two-dimensional CT lung region image includes:

[0018] The three-dimensional lung CT image is downsampled to obtain a compressed three-dimensional CT image.

[0019] Axial slicing is performed on the three-dimensional CT compressed image to obtain a two-dimensional CT lung region image corresponding to the three-dimensional CT compressed image.

[0020] Preferably, the prediction model includes a customized COVID-Net module, a dense Dense-Net module, a fast and efficient Efficient-Net module, a fused residual Inception-Net module, and a residual Res-Net module. The step of performing layered prediction based on the two-dimensional CT lung region image and outputting the first prediction result of each CT slice in the two-dimensional CT lung region image includes:

[0021] The customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fused residual Inception-Net module, and the residual Res-Net module are used to perform layered prediction on the two-dimensional CT lung region image, and the prediction results of each CT slice corresponding to each module in the two-dimensional CT lung region image are output.

[0022] The prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module are averaged, and the average value is used as the first prediction result of each CT slice.

[0023] Preferably, the step of predicting severe illness based on the first prediction result of each CT slice using a preset logistic regression model and outputting the corresponding severe illness weight includes:

[0024] The logistic regression model is used to perform weight analysis and calculation on the first prediction result of each CT slice, and the severe case prediction is performed based on the result of the weight analysis and calculation, and the corresponding severe case weight is output.

[0025] Preferably, after the step of predicting severe illness based on the first prediction result of each CT slice using a preset logistic regression model and outputting the corresponding severe illness weight, the method further includes:

[0026] If the predicted probability of severe illness corresponding to the severity weight is not less than a preset value, then the patient corresponding to the severity weight is determined to be a person at risk of severe illness.

[0027] If the predicted probability of severe illness corresponding to the severity weight is less than a preset value, then the patient corresponding to the severity weight is determined to be a population without risk of developing severe illness.

[0028] Furthermore, to achieve the above objectives, the present invention also provides a device for predicting the severity of pneumonia, the device comprising:

[0029] The acquisition module is used to acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images.

[0030] The layering module is used to perform layered prediction based on the two-dimensional CT lung region image through a prediction model, and output the first prediction result of each CT slice in the two-dimensional CT lung region image.

[0031] The critical care module is used to predict the severity of illness based on the first prediction result of each CT slice, using a preset logistic regression model, and output the corresponding severity weight.

[0032] The correction module is used to predict the three-dimensional lung CT images, obtain the corresponding pneumonia assessment value, and correct the pneumonia assessment value and the severity weight to determine the final prediction result.

[0033] In addition, to achieve the above objectives, the present invention also provides an electronic device, the electronic device comprising: a memory, a processor, and a pneumonia severity prediction program stored in the memory and executable on the processor, wherein when the pneumonia severity prediction program is executed by the processor, it implements the steps of the pneumonia severity prediction method as described above.

[0034] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a pneumonia severity prediction program, wherein the pneumonia severity prediction program, when executed by a processor, implements the steps of the pneumonia severity prediction method as described above.

[0035] This invention proposes a method, device, electronic device, and medium for predicting the severity of pneumonia. The method includes: acquiring three-dimensional lung CT images of a patient and preprocessing the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images; based on the two-dimensional CT lung region images, performing layered prediction using a prediction model and outputting a first prediction result for each CT slice in the two-dimensional CT lung region images; based on the first prediction result of each CT slice, performing severe case prediction using a pre-set logistic regression model and outputting corresponding severe case weights; predicting the three-dimensional lung CT images to obtain corresponding pneumonia assessment values, and correcting the pneumonia assessment values ​​and the severe case weights to determine the final prediction result; thereby using the prediction model to predict the severity indicators of each layer sequentially using CT layering; and using the logistic regression model to re-analyze and calculate the weights of different layers in the prediction results output by the prediction model to predict the severity of pneumonia; improving the accuracy of predicting the severity of pneumonia while enhancing the performance and medical interpretability of the machine learning model. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention;

[0037] Figure 2 This is a flowchart illustrating a method for predicting the severity of pneumonia according to the first embodiment of the present invention.

[0038] Figure 3 This is a schematic diagram illustrating the influence of different CT slice layers on the prediction effect provided in the first embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the overall analysis scheme for predicting severe pneumonia provided in the first embodiment of the present invention;

[0040] Figure 5 This is a flowchart illustrating a method for predicting the severity of pneumonia according to a second embodiment of the present invention.

[0041] Figure 6 This is a flowchart illustrating a method for predicting the severity of pneumonia according to a third embodiment of the present invention.

[0042] Figure 7 This is a flowchart illustrating a method for predicting the severity of pneumonia according to the fourth embodiment of the present invention.

[0043] Figure 8 This is a flowchart illustrating a method for predicting the severity of pneumonia according to the fifth embodiment of the present invention.

[0044] Figure 9 This is a schematic diagram illustrating the analysis results of the logistic regression model provided in the sixth embodiment of the present invention for different CT slice weights;

[0045] Figure 10 This is a schematic diagram of the functional modules of the first embodiment of the pneumonia severity prediction device of the present invention.

[0046] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0047] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0048] like Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0049] The electronic device in this embodiment of the invention can be a mobile terminal or a server device.

[0050] like Figure 1As shown, the device may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0051] Those skilled in the art will understand that Figure 1 The device structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0052] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a program for predicting the severity of pneumonia.

[0053] The operating system is a program that manages and controls electronic devices and software resources, and supports the operation of the network communication module, user interface module, pneumonia severity prediction program, and other programs or software; the network communication module is used to manage and control the network interface 1002; and the user interface module is used to manage and control the user interface 1003.

[0054] exist Figure 1 In the electronic device shown, the electronic device calls the pneumonia severity prediction program stored in the memory 1005 through the processor 1001, and performs the operations in the various embodiments of the pneumonia severity prediction method described below.

[0055] Based on the above hardware structure, an embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed.

[0056] This embodiment provides a method for predicting the severity of pneumonia, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the method for predicting the severity of pneumonia according to the present invention. The method for predicting the severity of pneumonia may include:

[0057] Step S10: Acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images.

[0058] Among them, three-dimensional lung CT (Computed Tomography) images are three-dimensional images obtained from chest CT scans of patients.

[0059] Preprocessing is used to remove irrelevant information in three-dimensional lung CT images, enhance the detectability of relevant information, and simplify the data processing process to the greatest extent possible; preprocessing may include image segmentation, image slice extraction, and other processing.

[0060] A slice refers to the image corresponding to a cross section in a three-dimensional image.

[0061] A slice can be a plane passing through a point inside a 3D image and parallel to the XY, YZ, and XZ planes, or it can be any plane passing through a point inside a 3D image and in any direction. By extracting slices, it is convenient to browse and analyze the internal structure of an image.

[0062] Image slicing extraction involves slicing a 3D image to extract its corresponding 2D image. For example, slicing a 240×240×240 pixel 3D image results in 240 layers of 2D images.

[0063] Two-dimensional CT lung region images can be two-dimensional images of lung slices obtained by slicing three-dimensional lung CT images.

[0064] It can acquire three-dimensional lung CT images of patients, and then perform image segmentation and image slice extraction on the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images.

[0065] Step S20: Based on the two-dimensional CT lung region image, perform layered prediction using a prediction model, and output the first prediction result for each CT slice in the two-dimensional CT lung region image.

[0066] The prediction model consists of five pre-trained two-dimensional convolutional neural network models, which are used to perform layered prediction based on the image of lung lobes in two-dimensional CT lung region images, and output the prediction results of each CT slice in the two-dimensional CT lung region image. The five different two-dimensional convolutional neural network models are the customized COVID-Net network model, the dense Dense-Net network model, the fast and efficient Efficient-Net network model, the fused residual Inception-Net network model, and the residual Res-Net network model.

[0067] The first prediction result is the prediction result determined based on the prediction results of five different two-dimensional convolutional neural network models in the prediction model.

[0068] The prediction results are obtained by five different two-dimensional convolutional neural network models performing hierarchical prediction based on the image of the upper lobe in the two-dimensional CT lung region image.

[0069] The infection process of the virus within an infected person follows a pattern, which is reflected in CT images of the lungs. Therefore, the differences in the appearance of different lung regions can help predict the severity of illness. Taking COVID-19 infection as an example, infected individuals initially show ground-glass opacities in the lower lobes of both lungs. As the disease progresses, these ground-glass opacities gradually transform into cobblestones and consolidation, and the patient's condition gradually improves. However, in severe cases, the lesions gradually spread to the upper lobes, worsening the condition. This indicates that the imaging findings of the upper lobes directly reflect the disease progression in patients with COVID-19-induced pneumonia. Therefore, the imaging manifestations of the upper lobes make a crucial contribution to predicting the severity of illness in patients.

[0070] After obtaining a two-dimensional CT lung region image, a prediction model can be used to perform layered prediction on the two-dimensional CT lung region image, and then the first prediction result of each CT slice in the two-dimensional CT lung region image can be output.

[0071] As an example, refer to Figure 3 , Figure 3 This diagram illustrates the impact of different CT slice layers on prediction results. Figure 3 The diagram shows the area under the curve (AUC) of different CNNs predicting different slices of data as input. A higher AUC indicates a better predictive model. Figure 3 The results show that the closer the slice is to the upper lung lobe of the input, the better the prediction model performs, which is consistent with the pathological development of pneumonia.

[0072] Step S30: Based on the first prediction result of each CT slice, a pre-set logistic regression model is used to predict the severity of the disease and output the corresponding severity weight.

[0073] Logistic regression is a pre-trained model that uses machine learning methods to perform weight analysis and calculation on the prediction results of the prediction model.

[0074] The severity weight is an indicator used to indicate whether a patient will develop severe pneumonia; the larger the severity weight value, the greater the probability that the patient corresponding to that severity weight will develop severe pneumonia.

[0075] After obtaining the first prediction result of each CT slice, the weight analysis of the first prediction result of each CT slice can be performed by a logistic regression model, and then the corresponding severe case weight can be obtained after the severe case prediction.

[0076] Furthermore, in one embodiment, after step S30, the method may further include the following steps:

[0077] If the predicted probability of severe illness corresponding to the severity weight is not less than a preset value, then the patient corresponding to the severity weight is determined to be a person at risk of severe illness.

[0078] If the predicted probability of severe illness corresponding to the severity weight is less than a preset value, then the patient corresponding to the severity weight is determined to be a population without risk of developing severe illness.

[0079] The preset value is a pre-set probability value used to predict whether a patient will develop severe or non-severe illness. As an example, the preset value can be 0.5. If the model predicts a probability of severe illness corresponding to the severity weight is greater than 0.5, the patient is classified as a person at risk of severe illness; otherwise, if it is less than 0.5, the patient is classified as a person without risk of developing severe illness.

[0080] In this embodiment, if the predicted probability of severe illness corresponding to the severity weight is not less than a preset value, the patient corresponding to the severity weight is determined to be in the high-risk group for severe illness; if the predicted probability of severe illness corresponding to the severity weight is less than the preset value, the patient corresponding to the severity weight is determined to be in the low-risk group for severe illness. Thus, the doctor can be prompted according to the severity weight and given the probability that the patient has a high risk of severe illness, thereby assisting medical staff in making relevant decisions.

[0081] Step S40: Predict the three-dimensional lung CT image to obtain the corresponding pneumonia assessment value, and correct the pneumonia assessment value and the severity weight to determine the final prediction result.

[0082] The pneumonia assessment value is a predicted value determined by combining the pathological characteristics of pneumonia with the patient's three-dimensional lung CT images. The higher the pneumonia assessment value, the greater the probability that the patient will be infected with pneumonia.

[0083] The final prediction result is determined after adjusting for the severity weight of the patient and the pneumonia assessment value. The final prediction result may include either severe pneumonia or non-severe pneumonia, and may also include the type of pneumonia.

[0084] By combining the pathological characteristics of pneumonia, the three-dimensional lung CT images of patients can be predicted to obtain the pneumonia assessment value corresponding to the three-dimensional lung CT images; then the pneumonia assessment value and the severity weight can be corrected to determine the final prediction result corresponding to the three-dimensional lung CT images.

[0085] As an example, refer to Figure 4 , Figure 4 This diagram illustrates the overall analysis scheme for predicting severe pneumonia. First, three-dimensional lung CT images of the patient are acquired and preprocessed to obtain corresponding two-dimensional CT images of the lung regions. These two-dimensional lung region images are then input into an ensemble learning module (i.e., a prediction model), which predicts the severity of the lung regions. The first prediction result is then input into a correction module based on image knowledge (i.e., a logistic regression model). Next, the logistic regression model adjusts the output weights based on imaging knowledge to predict the severity of the first prediction result, obtaining corresponding severity-related weights. Combining the pathological characteristics of pneumonia, the three-dimensional lung CT images are predicted to obtain corresponding pneumonia assessment values. Finally, the severity-related weights and pneumonia assessment values ​​are corrected to determine the final prediction result.

[0086] This embodiment acquires three-dimensional lung CT images of patients and preprocesses them to obtain corresponding two-dimensional CT lung region images. Based on the two-dimensional CT lung region images, a prediction model is used for stratified prediction, outputting the first prediction result for each CT slice in the two-dimensional CT lung region images. Based on the first prediction result of each CT slice, a pre-set logistic regression model is used for severe case prediction, outputting the corresponding severe case weight. The three-dimensional lung CT images are then predicted to obtain the corresponding pneumonia assessment value, and the pneumonia assessment value and the severe case weight are corrected to determine the final prediction result. Thus, the prediction model uses CT stratification to predict the severity indicators of each layer sequentially. The logistic regression model is used to re-analyze and calculate the weights of different layers in the prediction results output by the prediction model to predict the severity of pneumonia. This improves the accuracy of pneumonia severe case prediction and enhances the performance and medical interpretability of the machine learning model.

[0087] Based on the first embodiment, a second embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed, referring to... Figure 5 In this embodiment, before step S10, the method may further include the following steps:

[0088] Step A10: Obtain multiple different sample two-dimensional CT lung region images, and construct a training set from the sample two-dimensional CT lung region images for training a preset number of different two-dimensional convolutional neural network models.

[0089] The sample 2D CT lung region images are 2D images of lung slices from patients. These sample 2D CT lung region images can include images of severe pneumonia, non-severe pneumonia, and different types of pneumonia.

[0090] Two-dimensional convolutional neural network models (two-dimensional CNN models) can include five different convolutional neural network models: the custom COVID-Net network model, the dense Dense-Net network model, the fast and efficient Efficient-Net network model, the fused residual Inception-Net network model, and the residual Res-Net network model.

[0091] Each 2D CNN model uses a 7×7 convolutional module as the input to the first layer. Then, depending on the network structure of each 2D CNN model, either a 3×3 or a 1×1 convolutional module is selected for the parameter settings of subsequent convolutional layers. Furthermore, except for the output of the last layer which uses the Softmax function as the activation function, all other layers use the ReLU function as the activation function.

[0092] It can acquire sample 2D CT lung region images from multiple different patients, and then use these sample 2D CT lung region images as a basis to construct a training set for training five different convolutional neural network models.

[0093] During the construction of the training set, these sample 2D CT lung region images need to be preprocessed to ensure that the images extracted from the sample 2D CT lung region images have uniform size and format, which is to facilitate batch processing.

[0094] In practical applications, the more 2D CT lung region images in the training set, the more accurate the prediction results output by the trained prediction model will be.

[0095] Step A20: Obtain the sample annotation results corresponding to the two-dimensional CT lung region images of each sample.

[0096] The sample annotation results are the results of users pre-annotating sample 2D CT lung region images. The sample annotation results can include annotated images of severe pneumonia, non-severe pneumonia, and pneumonia types.

[0097] Obtain the sample annotation results of the user's annotation of the lung region in each sample 2D CT image.

[0098] The sample annotation results also need to be preprocessed to ensure that the size and format of each sample annotation result are consistent with the corresponding sample 2D CT lung region image, which is to facilitate batch processing.

[0099] Step A30: Use the sample 2D CT lung region images in the training set as input to the preset number of different 2D convolutional neural network models, use the sample annotation results as output to the preset number of different 2D convolutional neural network models, and iteratively train the preset number of different 2D convolutional neural network models until the accuracy of the validation set of the 2D convolutional neural network model reaches a stable level after iterative training, and then stop training to obtain the corresponding prediction model.

[0100] In some embodiments, the sample 2D CT lung region images and the corresponding sample annotation results of each sample 2D CT lung region image are sequentially input into five different convolutional neural network models, namely the customized COVID-Net network model, the dense Dense-Net network model, the fast and efficient Efficient-Net network model, the fused residual Inception-Net network model, and the residual Res-Net network model.

[0101] Two-dimensional CT images of lung regions were used as inputs to five different convolutional neural network (CNN) models, and the sample annotation results were used as outputs to the five different CNN models. Each of the five different CNN models was trained iteratively until the accuracy of the validation set of each 2D CNN model reached a stable value after iterative training, at which point the training was stopped and the corresponding prediction model was obtained.

[0102] In this embodiment, the sample 2D CT lung region images from the training set are used as input to a predetermined number of different 2D convolutional neural network models, and the sample annotation results are used as outputs of the predetermined number of different 2D convolutional neural network models. The predetermined number of different 2D convolutional neural network models are iteratively trained until the accuracy of the validation set of the 2D convolutional neural network model stabilizes after iterative training, at which point training stops, and the corresponding prediction model is obtained. Thus, by training five different 2D convolutional neural network models independently, the corresponding prediction model is obtained, thereby improving the prediction accuracy of the trained prediction model.

[0103] Based on the first and second embodiments, a third embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed. (Refer to...) Figure 6 In this embodiment, before step S10, the method may further include the following steps:

[0104] Step B10: Acquire multiple different sample 2D CT lung region images, and use the prediction model to perform layered prediction on the multiple different sample 2D CT lung region images to obtain the sample prediction results of different slices corresponding to each sample 2D CT lung region image.

[0105] Among them, the sample 2D CT lung region image is a 2D image of a slice of the patient's lungs.

[0106] As an example, multiple different samples of two-dimensional CT images of the lung region can include images of severe pneumonia, non-severe pneumonia, and different types of pneumonia.

[0107] It can acquire multiple different sample 2D CT lung region images, and then use a prediction model to perform layered prediction on multiple different sample 2D CT lung region images to obtain the sample prediction results of different slices corresponding to each sample 2D CT lung region image.

[0108] Step B20: Based on the sample prediction results of the different slices, the initial logistic regression model is trained and the severity of the illness is regressed to obtain the trained logistic regression model.

[0109] The initial logistic regression model is obtained by re-analyzing and recalculating the weights of different layers using machine learning methods.

[0110] The logistic regression model is an initial logistic regression model that uses machine learning methods to reanalyze and calculate the weights of different layers in the output of the prediction model.

[0111] After obtaining the sample prediction results of different slices, the sample prediction results of different slices can be trained by the initial logistic regression model, and the severity of the illness can be regressed to obtain the trained logistic regression model.

[0112] In this embodiment, multiple different sample 2D CT lung region images are acquired, and the prediction model is used to perform layered prediction on the multiple different sample 2D CT lung region images to obtain the sample prediction results of different slices corresponding to each sample 2D CT lung region image; based on the sample prediction results of different slices, the initial logistic regression model is trained and the severity of the illness is regressed to obtain the trained logistic regression model; thus, by using the first prediction results of different slices output by the prediction model to train the initial logistic regression model, the accuracy and stability of the trained logistic regression model are improved.

[0113] Based on the first embodiment, a fourth embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed. (Refer to...) Figure 7 In this embodiment, step S10 may further include the following sub-steps:

[0114] Sub-step S11: The three-dimensional lung CT image is downsampled to obtain a compressed three-dimensional CT image after compression of the three-dimensional lung CT image.

[0115] Downsampling involves reducing the size of the image to be processed so that the processed image fits the size of the display area, and outputting a thumbnail of the image to be processed.

[0116] After obtaining three-dimensional lung CT images, the images can be standardized to ensure uniform image size, facilitating batch processing. Then, the standardized three-dimensional lung CT images can be compressed into 240×240×240 pixel three-dimensional CT compressed images by downsampling.

[0117] Sub-step S12 involves axially slicing the three-dimensional CT compressed image to obtain a two-dimensional CT lung region image corresponding to the three-dimensional CT compressed image.

[0118] After obtaining a 240×240×240 pixel three-dimensional CT compressed image, the three-dimensional CT compressed image can be axially sliced ​​into 240 layers of two-dimensional images. Then, the middle 120 layers, including the lungs, are taken to obtain the corresponding two-dimensional CT lung region image.

[0119] In this embodiment, by downsampling the three-dimensional lung CT image, a compressed three-dimensional CT image is obtained; the compressed three-dimensional CT image is then axially sliced ​​to obtain a corresponding two-dimensional CT lung region image; thus, by preprocessing the three-dimensional lung CT image, the corresponding two-dimensional CT lung region image is obtained, thereby improving the accuracy of predicting severe illness.

[0120] Based on the first and second embodiments, a third embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed. (Refer to...) Figure 8 In this embodiment, step S20 may include the following sub-steps:

[0121] Sub-step S21 involves performing layered predictions on the two-dimensional CT lung region image using the customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fused residual Inception-Net module, and the residual Res-Net module, and outputting the prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module.

[0122] In the case where the prediction model includes a customized COVID-Net module, a dense Dense-Net module, a fast and efficient Efficient-Net module, a fusion residual Inception-Net module, and a residual Res-Net module, after inputting the above-mentioned intermediate 120 layers of 2D CT lung region images, including the lung portion, into the prediction model, the customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fusion residual Inception-Net module, and the residual Res-Net module respectively perform layered prediction on the 2D CT lung region images to obtain the prediction results of each CT slice in the 2D CT lung region images corresponding to each module.

[0123] Sub-step S22: Take the average value of the prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module, and use the average value as the first prediction result of each CT slice.

[0124] Specifically, after obtaining the prediction results of each CT slice in the two-dimensional CT lung region image output by the above modules, the prediction results of each CT slice in the two-dimensional CT lung region image are averaged, and the average of the results of each slice is used as the first prediction result of each CT slice.

[0125] In this embodiment, the customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fused residual Inception-Net module, and the residual Res-Net module are used to perform layered prediction on the two-dimensional CT lung region image, and the prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module are output. The average value of the prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module is taken, and the average value is used as the first prediction result of each CT slice. Thus, by performing layered prediction on the two-dimensional CT lung region image through the prediction model, the first prediction result of each CT slice in the corresponding two-dimensional CT lung region image is obtained, which can more accurately predict the conversion of pneumonia patients to severe illness.

[0126] Based on the first embodiment, a sixth embodiment of the method for predicting the severity of pneumonia according to the present invention is proposed. In this embodiment, step S30 may include the following sub-steps:

[0127] Sub-step S31 involves performing weight analysis and calculation on the first prediction result of each CT slice using the logistic regression model, and then performing severe case prediction based on the results of the weight analysis and calculation, outputting the corresponding severe case weight.

[0128] After obtaining the first prediction result of each CT slice, the first prediction result of each CT slice can be input into the logistic regression model. The logistic regression model is used to perform weight analysis and calculation on the first prediction result of each CT slice, and the severe case prediction is performed based on the result of the weight analysis and calculation, and the corresponding severe case weight is output.

[0129] As an example, refer to Figure 9 , Figure 9 This is a schematic diagram illustrating the analysis results of the logistic regression model with different CT slice weights. Figure 9 It can be seen that the results from slices 115-120 contribute the most to the final prediction. This is consistent with the conclusions of the previous pathological analysis, namely that for critically ill patients, the lesions gradually spread to the upper lobes of the lungs and worsen the condition. Therefore, we selected these slices close to the upper lobes as input to the logistic regression model, and after weight analysis, output the severity weight corresponding to each CT slice.

[0130] In this embodiment, the first prediction result of each CT slice is weighted using the logistic regression model, and the severity prediction is made based on the weighted analysis result, outputting the corresponding severity weight; thereby improving the accuracy of severity prediction by re-analyzing and calculating the weights of different layers using machine learning methods.

[0131] This invention also provides a device for predicting the severity of pneumonia. (Refer to...) Figure 10 The pneumonia severity prediction device of the present invention may include:

[0132] The acquisition module 10 is used to acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images.

[0133] The layering module 20 is used to perform layered prediction based on the two-dimensional CT lung region image through a prediction model, and output the first prediction result of each CT slice in the two-dimensional CT lung region image.

[0134] The critical care module 30 is used to predict the severity of illness based on the first prediction result of each CT slice by using a preset logistic regression model and output the corresponding severity weight.

[0135] The correction module 40 is used to predict the three-dimensional lung CT image, obtain the corresponding pneumonia assessment value, and correct the pneumonia assessment value and the severity weight to determine the final prediction result.

[0136] In addition, the present invention provides a computer-readable storage medium storing a pneumonia severity prediction program thereon, which, when executed by a processor, implements the steps of the pneumonia severity prediction method as described above.

[0137] The method implemented when the pneumonia severity prediction program running on the processor is executed can be referred to in various embodiments of the pneumonia severity prediction method of the present invention, and will not be repeated here.

[0138] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0139] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause an electronic device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0141] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting the severity of pneumonia, characterized in that, The method includes: Acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images; Based on the two-dimensional CT lung region image, a prediction model is used for layered prediction, outputting the first prediction result for each CT slice in the two-dimensional CT lung region image. The prediction model includes a customized COVID-Net module, a dense Dense-Net module, a fast and efficient Efficient-Net module, a fused residual Inception-Net module, and a residual Res-Net module. The customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fused residual Inception-Net module, and the residual Res-Net module respectively perform layered prediction on the two-dimensional CT lung region image and output the prediction result for each CT slice in the two-dimensional CT lung region image corresponding to each module. The average value of the prediction results for each CT slice in the two-dimensional CT lung region image corresponding to each module is taken, and the average value is used as the first prediction result for each CT slice. Based on the first prediction result of each CT slice, a pre-set logistic regression model is used to predict the severity of the disease and output the corresponding severity weight. Specifically, the logistic regression model is used to perform weight analysis and calculation on the first prediction result of each CT slice, and the severity of the disease is predicted based on the results of the weight analysis and calculation, and the corresponding severity weight is output. The three-dimensional lung CT images are used to predict the corresponding pneumonia assessment value, and the pneumonia assessment value and the severity weight are corrected to determine the final prediction result. After the step of predicting severe illness based on the first prediction result of each CT slice using a preset logistic regression model and outputting the corresponding severe illness weight, the method further includes: If the predicted probability of severe illness corresponding to the severity weight is not less than a preset value, then the patient corresponding to the severity weight is determined to be a person at risk of severe illness; if the predicted probability of severe illness corresponding to the severity weight is less than a preset value, then the patient corresponding to the severity weight is determined to be a person without risk of developing severe illness.

2. The method as described in claim 1, characterized in that, Before the step of acquiring the patient's three-dimensional lung CT images and preprocessing the three-dimensional lung CT images to obtain the corresponding three-dimensional CT compressed images, the method further includes: Multiple different sample two-dimensional CT lung region images are acquired, and the sample two-dimensional CT lung region images are used to construct a training set for training a preset number of different two-dimensional convolutional neural network models; Obtain the sample annotation results corresponding to the two-dimensional CT lung region images of each sample; The sample 2D CT lung region images in the training set are used as inputs to a predetermined number of different 2D convolutional neural network models. The sample annotation results are used as outputs to the predetermined number of different 2D convolutional neural network models. The predetermined number of different 2D convolutional neural network models are iteratively trained until the accuracy of the validation set of the 2D convolutional neural network model reaches a stable level after iterative training, at which point training stops, and the corresponding prediction model is obtained.

3. The method as described in claim 1, characterized in that, Before the step of acquiring the patient's three-dimensional lung CT images and preprocessing the three-dimensional lung CT images to obtain the corresponding two-dimensional CT lung region images, the method further includes: Multiple different sample 2D CT lung region images are acquired, and the prediction model is used to perform layered prediction on the multiple different sample 2D CT lung region images to obtain the sample prediction results of different layer slices corresponding to each sample 2D CT lung region image. Based on the sample prediction results of the different slices, the initial logistic regression model is trained and the severity of the illness is regressed to obtain the trained logistic regression model.

4. The method as described in claim 1, characterized in that, The step of preprocessing the three-dimensional lung CT image to obtain the corresponding two-dimensional CT lung region image includes: The three-dimensional lung CT image is downsampled to obtain a compressed three-dimensional CT image. Axial slicing is performed on the three-dimensional CT compressed image to obtain a two-dimensional CT lung region image corresponding to the three-dimensional CT compressed image.

5. A device for predicting the severity of pneumonia, characterized in that, The device includes: The acquisition module is used to acquire three-dimensional lung CT images of the patient and preprocess the three-dimensional lung CT images to obtain corresponding two-dimensional CT lung region images. A layering module is used to perform layered prediction based on the two-dimensional CT lung region image using a prediction model, and output the first prediction result of each CT slice in the two-dimensional CT lung region image. The prediction model includes a customized COVID-Net module, a dense Dense-Net module, a fast and efficient Efficient-Net module, a fused residual Inception-Net module, and a residual Res-Net module. The customized COVID-Net module, the dense Dense-Net module, the fast and efficient Efficient-Net module, the fused residual Inception-Net module, and the residual Res-Net module respectively perform layered prediction on the two-dimensional CT lung region image and output the prediction result of each CT slice in the two-dimensional CT lung region image corresponding to each module. The average value of the prediction results of each CT slice in the two-dimensional CT lung region image corresponding to each module is taken, and the average value is used as the first prediction result of each CT slice. The critical care module is used to predict the severity of a disease based on the first prediction result of each CT slice using a preset logistic regression model, and output the corresponding severity weight. Specifically, the logistic regression model is used to perform weight analysis and calculation on the first prediction result of each CT slice, and the severity prediction is performed based on the result of the weight analysis and calculation, and the corresponding severity weight is output. The correction module is used to predict the three-dimensional lung CT images, obtain the corresponding pneumonia assessment value, and correct the pneumonia assessment value and the severity weight to determine the final prediction result. The critical care module is further configured to determine the patient corresponding to the critical care weight as a critical care risk group if the predicted probability of critical care corresponding to the critical care weight is not less than a preset value; and to determine the patient corresponding to the critical care weight as a non-critical care risk group if the predicted probability of critical care corresponding to the critical care weight is less than a preset value.

6. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 4.

Images