Pelvic fracture automatic segmentation method and system based on convolutional neural network

Abstract

The invention discloses a pelvic fracture automatic segmentation method based on a convolutional neural network, and a deep learning technology, especially the convolutional neural network (CNN), can identify and segment a pelvic fracture area more accurately. Therefore, a doctor can quickly and accurately know the specific position and severity of the fracture, so that the accuracy and reliability of diagnosis are improved, the time for the doctor to manually analyze the image is greatly shortened by an automatic processing flow, a patient can obtain a diagnosis result in a shorter time, and the method is particularly important for treatment decision under an emergency condition, and is suitable for popularization and application. According to the method, a set of standardized fracture recognition and segmentation algorithm is formed by learning a large amount of CT image data, and the misdiagnosis rate caused by experience difference of doctors can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical image reconstruction, and in particular to a convolutional neural network-based pelvic fracture automatic segmentation method and system. Background Art

[0002] Pelvic fracture is a serious damage and deformation of the pelvis caused by external factors. It is very harmful to the human body and is accompanied by severe pain, difficulty in lower limb movement, and in severe cases shock and syncope. To diagnose a pelvic fracture, the patient needs to undergo an X-ray examination, and a CT scan is required for a further accurate diagnosis. The doctor manually reconstructs the anatomical structure and fracture position relationship of each part of the patient's pelvic fracture in three dimensions to determine the type of pelvic fracture and determine the surgical plan.

[0003] Traditional pelvic fracture methods rely on manual drawing by experienced 3D reconstruction personnel, which is time-consuming and labor-intensive, prone to errors, and difficult to guarantee accuracy.

[0004] In summary, a pelvic fracture automatic segmentation method and system based on convolutional neural network is needed to solve the shortcomings of the existing technology. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method and system for automatic segmentation of pelvic fractures based on convolutional neural networks, aiming to solve the above problems.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for automatic segmentation of pelvic fractures based on a convolutional neural network, comprising the following steps: Step S1: Collect CT image data of patients with pelvic fractures; classify the data according to the location of the pelvic fracture; Step S2: training the convolutional neural network model using CT image data; Step S3: Optimize the convolutional neural network model training process by adjusting the learning rate and iteration number parameters until the training is completed; Step S4: using the trained model to automatically segment new pelvic fracture CT image data; Step S5: Determine the fracture type based on the segmentation results and formulate a preliminary fracture surgery plan.

[0007] Optionally, in step S1, the collected CT image data is processed: The CT image data is imported into the data preprocessing software, the fractured pelvic data is segmented, and the comminuted or open pelvic fracture parts requiring surgery are drawn out in sequence.

[0008] Optionally, the data classification in step S1 includes but is not limited to the healthy coccyx part, the fractured coccyx part, the healthy left pelvic part, the fractured left pelvic part, the healthy right pelvic part and the fractured right pelvic part, and the data classification is numbered.

[0009] Optionally, the convolutional neural network model training in step S2 includes image preprocessing and neural network processing. Image preprocessing is: Step A1: Format the dataset as required, with each case corresponding to a label file; Step A2: Run the preprocessing script, which runs commands for preprocessing and creates processing data and plan files; Step A3: After preprocessing is completed, check whether the subfolders and plan files in the output directory are complete.

[0010] Optionally, the neural network processing is: Step B1: Check the training environment and configuration, and adjust the version and settings as needed; Step B2: Use the output directory for training; Step B3: Monitor the training process and regularly check the loss value and performance indicators on the validation set; Step B4: Perform new data prediction or fine-tune the model. After training is completed, the convolutional neural network model is saved and the output directory is updated. The model is evaluated using the test dataset.

[0011] Optionally, the parameter optimization in step S3 is performed in the following manner: Step S31: Initialize the weights and biases of the model, using the pre-trained weights as initial values; Step S32: define the loss function, select a suitable loss function, and set the learning rate; Step S33: input training data, calculate the predicted segmentation mask through the encoder, bottleneck layer, decoder and output layer of the model, and use the defined loss function to calculate the loss between the predicted value and the true label; Step S34: Calculate the gradient of the loss function with respect to the model parameters, and use the gradient descent method to update the model parameters until the set stop condition is reached.

[0012] Optionally, the calculation formula of the gradient descent method is: , In the formula, is the parameter of the model at the next time step of time step t, are the parameters of the model at time step t, is the learning rate, is the gradient of the loss function J with respect to the parameter ω under the current parameters.

[0013] Optionally, the learning rate adjustment strategy includes manual, learning rate decay and learning rate search. Manual adjustment is done by initially setting a larger value and gradually fine-tuning it based on training feedback; Learning rate decay reduces the learning rate by a fixed ratio or dynamically based on performance indicators; Learning rate search optimizes the training process by testing a range of learning rates and cyclically varying the learning rate.

[0014] A pelvic fracture automatic segmentation system based on convolutional neural network, which adopts a pelvic fracture automatic segmentation method based on convolutional neural network, including a data collection module, a data collection module, a data annotation module, a model training module, an automatic segmentation module and a result output module. A data collection module, used for acquiring and preprocessing CT image data of patients with pelvic fractures; Data labeling module, used to classify and label data according to fracture sites; Model training module, used to train data based on convolutional neural networks and optimize model parameters; An automatic segmentation module is used to automatically segment new pelvic fracture CT image data using the trained model; The result output module is used to display the segmentation results and assist doctors in determining the fracture type and formulating surgical plans.

[0015] Optionally, the system also includes a model evaluation module for evaluating the segmentation performance of the model.

[0016] Beneficial effects of the present invention: 1. In the present invention, deep learning technology, especially convolutional neural network (CNN), can more accurately identify and segment the pelvic fracture area. This helps doctors quickly and accurately understand the specific location and severity of the fracture, thereby improving the accuracy and reliability of diagnosis. The automated processing flow greatly reduces the time for doctors to manually analyze images, allowing patients to get diagnosis results in a shorter time, which is particularly important for treatment decisions in emergency situations. This method forms a set of standardized fracture recognition and segmentation algorithms by learning a large amount of CT image data, which helps to reduce the misdiagnosis rate caused by differences in doctor experience.

[0017] 2. In the present invention, based on the results of automatic segmentation, the system can provide preliminary surgical plan recommendations for different fracture types, helping doctors to customize personalized treatment plans based on the specific conditions of the patients. The application of automated tools reduces the workload of doctors, allowing limited medical resources to serve more patients more efficiently.

[0018] 3. In the present invention, the accumulated large-scale, high-quality pelvic fracture dataset not only helps to improve the existing model, but also provides valuable data support for future medical research. The automatic segmentation system based on the cloud platform can enable medical institutions in remote areas to enjoy high-level medical diagnosis services, and promote the sinking of high-quality medical resources. By continuously adding new data and adjusting model parameters, the accuracy and generalization ability of the model can be continuously improved to ensure long-term optimal performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a method.

[0020] Figure 2 This is a diagram of a right pelvic fracture of the present invention.

[0021] Figure 3 This is a healthy side view of the right pelvis of the present invention.

[0022] Figure 4 It is a healthy side view of the left pelvis of the present invention.

[0023] Figure 5 This is a healthy side view of the coccyx of the present invention.

[0024] Figure 6 This is a partial identification diagram of the healthy coccyx of the present invention.

[0025] Figure 7 This is a partial identification diagram of a fractured coccyx according to the present invention.

[0026] Figure 8 This is a three-dimensional identification map of the pelvic part of the present invention.

[0027] Fig. 9 This is a model training curve diagram of the present invention. DETAILED DESCRIPTION

[0028] In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] like Figures 1 to 9 As shown, a pelvic fracture automatic segmentation method based on convolutional neural network includes the following steps: Step S1: Collect CT image data of patients with pelvic fractures; classify the data according to the location of the pelvic fracture; Step S2: training the convolutional neural network model using CT image data; Step S3: Optimize the convolutional neural network model training process by adjusting the learning rate and iteration number parameters until the training is completed; Step S4: using the trained model to automatically segment new pelvic fracture CT image data; Step S5: Determine the fracture type based on the segmentation results and formulate a preliminary fracture surgery plan.

[0030] In step S1, the collected CT image data is processed: The CT image data is imported into the data preprocessing software, the fractured pelvic data is segmented, and the comminuted or open pelvic fracture parts requiring surgery are drawn out in sequence.

[0031] The data classification in step S1 includes but is not limited to the healthy coccyx part, the fractured coccyx part, the healthy left pelvic part, the fractured left pelvic part, the healthy right pelvic part and the fractured right pelvic part, and the data classification is numbered.

[0032] The convolutional neural network model training in step S2 includes image preprocessing and neural network processing. Image preprocessing is: Step A1: Format the dataset as required, with each case corresponding to a label file; Step A2: Run the preprocessing script, which runs commands for preprocessing and creates processing data and plan files; Step A3: After preprocessing is completed, check whether the subfolders and plan files in the output directory are complete.

[0033] The neural network process is: Step B1: Check the training environment and configuration, and adjust the version and settings as needed; Step B2: Use the output directory for training; Step B3: Monitor the training process and regularly check the loss value and performance indicators on the validation set; Step B4: Perform new data prediction or fine-tune the model. After training is completed, the convolutional neural network model is saved and the output directory is updated. The model is evaluated using the test dataset.

[0034] Optionally, the parameter optimization in step S3 is performed in the following manner: Step S31: Initialize the weights and biases of the model, using the pre-trained weights as initial values; Step S32: define the loss function, select a suitable loss function, and set the learning rate; Step S33: input training data, calculate the predicted segmentation mask through the encoder, bottleneck layer, decoder and output layer of the model, and use the defined loss function to calculate the loss between the predicted value and the true label; Step S34: Calculate the gradient of the loss function with respect to the model parameters, and use the gradient descent method to update the model parameters until the set stop condition is reached.

[0035] The calculation formula of the gradient descent method is: , In the formula, is the parameter of the model at the next time step of time step t, are the parameters of the model at time step t, is the learning rate, is the gradient of the loss function J with respect to the parameter ω under the current parameters.

[0036] Learning rate adjustment strategies include manual, learning rate decay, and learning rate search. Manual adjustment is done by initially setting a larger value and gradually fine-tuning it based on training feedback; Learning rate decay reduces the learning rate by a fixed ratio or dynamically based on performance indicators; Learning rate search optimizes the training process by testing a range of learning rates and cyclically varying the learning rate.

[0037] A pelvic fracture automatic segmentation system based on convolutional neural network, which adopts a pelvic fracture automatic segmentation method based on convolutional neural network, including a data collection module, a data collection module, a data annotation module, a model training module, an automatic segmentation module and a result output module. A data collection module, used for acquiring and preprocessing CT image data of patients with pelvic fractures; Data labeling module, used to classify and label data according to fracture sites; Model training module, used to train data based on convolutional neural networks and optimize model parameters; An automatic segmentation module is used to automatically segment new pelvic fracture CT image data using the trained model; The result output module is used to display the segmentation results and assist doctors in determining the fracture type and formulating surgical plans. A model evaluation module is also included to evaluate the segmentation performance of the model.

[0038] The training data source of the present invention is provided by the hospital. The hospital removes sensitive information from the CT data and gives it to the data drawing and annotation personnel, who then produce the training data. The id of the established DataBase is 600, the project name is Dataset600_CT_PelvicFrac150, and the number of training data sets is 100, of which 80 are for training and 20 are for verification.

[0039] Data creation method: Import CT data into miniics, manually segment the fractured pelvis data, and draw all comminuted or open pelvic fractures that require surgery in sequence, ignoring minor fractures.

[0040] It is divided into: the healthy coccyx part; the fractured coccyx part; the healthy left pelvic part; the fractured left pelvic part; the healthy right pelvic part; the fractured right pelvic part; and the background color part.

[0041] For training, you need to set the data number, the number is as follows:

[0042] The training process is 1. Image preprocessing 2. Neural network processing.

[0043] Image preprocessing: Use nnUNet_plan_and_preprocess to create subfolders with the preprocessed data for the 2D U-Net and all applicable 3D U-Nets. It will also create "plan" files (with ending.pkl) for both 2D and 3D configurations. These files contain the generated segmentation pipeline configurations and will be read by nnUNetTrainer.

[0044] Neural Network Processing: The neural network model used for training is the V2 version of nnUnet, and the training adopts the Full3D mode, without using the 2D and 2D3D cross mode. After many experiments, the optimal initial learning rate parameter of training is adjusted to 0.02, each round is 250 times, and the first training round is 1000 rounds.

[0045] If the learning rate is set too large, the model parameters may cross the optimal solution in each iteration, resulting in unstable oscillation. On the contrary, if the learning rate is set too small, the model will converge to the optimal solution very slowly, and may not be able to find the global optimal solution due to the small step size.

[0046] The parameter optimization process was carried out through repeated tests, and it was finally determined that the learning rate was adjusted to 0.02 to ensure the optimal learning efficiency.

[0047] The present invention uses deep learning technology, especially convolutional neural network (CNN), to more accurately identify and segment the pelvic fracture area. This helps doctors quickly and accurately understand the specific location and severity of the fracture, thereby improving the accuracy and reliability of diagnosis. The automated processing flow greatly reduces the time doctors spend manually analyzing images, allowing patients to get diagnosis results in a shorter time, which is particularly important for treatment decisions in emergency situations. This method forms a set of standardized fracture identification and segmentation algorithms by learning from a large amount of CT image data, which helps reduce the misdiagnosis rate caused by differences in doctor experience.

[0048] Based on the results of automatic segmentation, the system can provide preliminary surgical plan recommendations for different fracture types, helping doctors customize personalized treatment plans based on the patient's specific circumstances. The application of automated tools reduces the workload of doctors and allows limited medical resources to serve more patients more efficiently.

[0049] The accumulated large-scale, high-quality pelvic fracture dataset not only helps to improve the existing models, but also provides valuable data support for future medical research. The automatic segmentation system based on the cloud platform can enable medical institutions in remote areas to enjoy high-level medical diagnosis services, and promote the sinking of high-quality medical resources. By continuously adding new data and adjusting model parameters, the accuracy and generalization ability of the model can be continuously improved to ensure long-term optimal performance.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent substitution or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatic segmentation of pelvic fractures based on convolutional neural networks, characterized in that: The following steps are involved: Step S1: Collect CT image data of patients with pelvic fractures; classify the data according to the location of the pelvic fracture; Step S2: training the convolutional neural network model using CT image data; Step S3: Optimize the convolutional neural network model training process by adjusting the learning rate and iteration number parameters until the training is completed; Step S4: using the trained model to automatically segment new pelvic fracture CT image data; Step S5: Determine the fracture type based on the segmentation results and formulate a preliminary fracture surgery plan.

2. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 1, characterized in that: In step S1, the collected CT image data is processed: The CT image data is imported into the data preprocessing software, the fractured pelvic data is segmented, and the comminuted or open pelvic fracture parts requiring surgery are drawn out in sequence.

3. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 1, characterized in that: The data classification in step S1 includes but is not limited to the healthy coccyx part, the fractured coccyx part, the healthy left pelvic part, the fractured left pelvic part, the healthy right pelvic part and the fractured right pelvic part, and the data classification is numbered.

4. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 1, characterized in that: The convolutional neural network model training in step S2 includes image preprocessing and neural network processing. Image preprocessing is: Step A1: Format the dataset as required, with each case corresponding to a label file; Step A2: Run the preprocessing script, which runs commands for preprocessing and creates processing data and plan files; Step A3: After preprocessing is completed, check whether the subfolders and plan files in the output directory are complete.

5. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 4, characterized in that: The neural network processing is: Step B1: Check the training environment and configuration, and adjust the version and settings as needed; Step B2: Use the output directory for training; Step B3: Monitor the training process and regularly check the loss value and performance indicators on the validation set; Step B4: Perform new data prediction or fine-tune the model. After training is completed, the convolutional neural network model is saved and the output directory is updated. The model is evaluated using the test dataset.

6. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 1, characterized in that: The parameter optimization in step S3 is carried out in the following manner: Step S31: Initialize the weights and biases of the model, using the pre-trained weights as initial values; Step S32: define the loss function, select a suitable loss function, and set the learning rate; Step S33: input training data, calculate the predicted segmentation mask through the encoder, bottleneck layer, decoder and output layer of the model, and use the defined loss function to calculate the loss between the predicted value and the true label; Step S34: Calculate the gradient of the loss function with respect to the model parameters, and use the gradient descent method to update the model parameters until the set stop condition is reached.

7. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 6, characterized in that: The calculation formula of the gradient descent method is: ， In the formula, is the parameter of the model at the next time step of time step t, are the parameters of the model at time step t, is the learning rate, is the gradient of the loss function J with respect to the parameter ω under the current parameters.

8. The method for automatic segmentation of pelvic fractures based on convolutional neural network according to claim 6, characterized in that: The learning rate adjustment strategies include manual, learning rate decay and learning rate search. Manual adjustment is done by initially setting a larger value and gradually fine-tuning it based on training feedback; Learning rate decay reduces the learning rate by a fixed ratio or dynamically based on performance indicators; Learning rate search optimizes the training process by testing a range of learning rates and cyclically varying the learning rate.

9. A pelvic fracture automatic segmentation system based on convolutional neural network, using the pelvic fracture automatic segmentation method based on convolutional neural network as claimed in claim 1, characterized in that: It includes data collection module, data collection module, data annotation module, model training module, automatic segmentation module and result output module. A data collection module, used for acquiring and preprocessing CT image data of patients with pelvic fractures; Data labeling module, used to classify and label data according to fracture sites; Model training module, used to train data based on convolutional neural networks and optimize model parameters; An automatic segmentation module is used to automatically segment new pelvic fracture CT image data using the trained model; The result output module is used to display the segmentation results and assist doctors in determining the fracture type and formulating surgical plans.

10. The automatic pelvic fracture segmentation system based on convolutional neural network according to claim 9, characterized in that: The system also includes a model evaluation module for evaluating the segmentation performance of the model.

Images