Image segmentation method for evaluating hepatocellular carcinoma neutron therapy dose

Abstract

The invention discloses an image segmentation method for evaluating a hepatocellular carcinoma neutron therapy dose, and relates to the technical field of neutron therapy. A 3D U-Net GAN network model is trained through a medical image data set; a generator model and a discriminator model in a 3D U-Net GAN network model are alternately trained by using a real image and a false image randomly generated by the generator model, so that the trained generator model can generate a segmentation result which is more accurate and rich in details for a medical image; the trained discriminator model can more accurately evaluate the authenticity of the output of the generator model, and the trained network can accurately reflect the anatomical structure of the patient, the distribution condition of 10B in the body and the radionuclide dynamics condition when segmenting the medical image, so that the information in the medical image can be accurately and fully displayed, and the medical image segmentation efficiency can be improved. The application is convenient.

Description

Technical Field

[0001] The present invention relates to the technical field of neutron therapy, and particularly to an image segmentation method, device, computer device and storage medium for evaluating the neutron therapy dose of hepatocellular carcinoma. Background Art

[0002] Boron neutron capture therapy (BNCT) utilizes the 10 B(n, α) 7 Li nuclear reaction occurring in cancer cells to release energy to kill cancer cells; in this process, α particles with extremely high linear energy transfer (LET) and 7 Li recoil nuclei are generated, with average energies of 190 keV / μm and 160 keV / μm respectively, which can damage diseased cells in the human body while having extremely short ranges.

[0003] Hepatocellular carcinoma (HCC) is an invasive primary liver cancer, accounting for more than 80% of primary liver cancers worldwide, ranking 6th in the global incidence of malignant tumors in 2018 and 4th in the causes of tumor deaths; in China, hepatocellular carcinoma ranks 4th in the malignant incidence in 2015 and 3rd in the causes of tumor deaths; common methods for treating hepatocellular carcinoma include, for example, surgical treatment, liver transplantation, local treatment (radiofrequency ablation, microwave treatment, and ethanol injection treatment), transarterial chemoembolization, radiotherapy, and supportive treatment, etc.; compared with traditional therapies, neutron therapy can more precisely damage cancer cells due to its relatively short range, thereby reducing damage to surrounding normal tissues.

[0004] However, since it is impossible to directly measure the dosimetric parameters inside the human body, currently, it is usually based on manual division of tissues and organs in medical images to support subsequent analysis, but this method is difficult to reflect the anatomical structures of different patients themselves, 10 the distribution of Summary of the Invention

[0005] Embodiments of the present invention provide an image segmentation method for evaluating the neutron therapy dose of hepatocellular carcinoma, which can solve the problem in the prior art that this method is difficult to reflect the anatomical structures of different patients themselves, 10 the distribution of

[0006] An embodiment of the present invention provides an image segmentation method for evaluating the dose of neutron therapy in hepatocellular carcinoma, including the following steps:

[0007] Obtain a medical image dataset of the hepatocellular carcinoma region;

[0008] Construct a 3D U-Net generator model and a 3D CNN discriminator model, and couple the 3D U-Net generator model and the 3D CNN discriminator model into a 3D U-Net GAN network model;

[0009] Use the medical image dataset to train the 3D U-Net GAN network model. Take the real images in the medical image dataset and the fake images randomly generated by the generator model as the inputs of the 3D U-Net GAN network model. Obtain a preliminary segmentation result through forward propagation in the 3D U-Net GAN network model. Obtain the losses of the generator model and the discriminator model in the 3D U-Net GAN network model according to the loss function. Update the parameters of the 3D U-Net GAN network model through backpropagation in the 3D U-Net GAN network model according to the preliminary segmentation result and the losses of the generator model and the discriminator model. At the same time, fix the discriminator model to train the generator model, and fix the generator model to train the discriminator model to obtain the trained 3D U-Net GAN network model;

[0010] Input the medical image into the trained 3D U-Net GAN network model. Use the generator model in the trained 3D U-Net GAN network model to extract the features of the liver part and the tumor part in the medical image, and respectively obtain the structural feature maps including three-dimensional stereo structures and morphological structures and the radioactive element feature maps including radioactive element distributions and radionuclides of the liver part and the tumor part. Use the discriminator model to screen and remove the false feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part, and respectively sum up the remaining true feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part, and perform imageization to respectively obtain the segmentation images of the liver part and the tumor part. Connect the segmentation images of the liver part and the tumor part to obtain the segmented image.

[0011] Preferably, the acquisition of the medical image dataset includes:

[0012] Obtain a medical image of the hepatocellular carcinoma region;

[0013] Perform Gaussian filtering technology, median filtering technology, bilateral filtering technology, histogram equalization, adaptive histogram equalization, contrast stretching, and geometric transformation processing on the medical image to obtain a medical image dataset of the hepatocellular carcinoma region;

[0014] Randomly rotate, horizontally flip, vertically flip, randomly crop, randomly adjust brightness or randomly adjust contrast of the medical image dataset to enhance the data in the medical image dataset.

[0015] Preferably, the construction of the 3D U-Net generator model and the 3D CNN discriminator model includes:

[0016] The 3D U-Net generator model adds multiple convolutional layers and pooling layers between the encoder and decoder in the model to adapt to 3D data;

[0017] The 3D U-Net generator model generates realistic samples from random noise or other inputs, and generates new samples similar to real samples by learning the distribution and features of the sample data;

[0018] The 3D CNN discriminator model uses different network structures to distinguish authenticity from real samples and new samples similar to real samples generated.

[0019] Preferably, the forward propagation algorithm includes:

[0020] The forward propagation in the generator part is as follows: The input data is passed through a series of neural network layers to gradually increase the dimension of the input data to generate an image. The generator usually consists of multiple layers, including fully connected layers or convolutional layers, activation functions, upsampling layers. In some variants, the generator may contain skip connections. Finally, the generator outputs the generated image through an activation function;

[0021] The forward propagation in the discriminator part is as follows: It receives both the real image and the fake image generated by the generator as inputs, extracts features and classifies them through a series of neural network layers, and gradually reduces the spatial dimension of the image through a pooling layer or strided convolution to extract more abstract features. The final output is the discriminator's prediction result on whether the input image is real or fake.

[0022] Preferably, fix the discriminator model to train the generator model, and then fix the generator model to train the discriminator model, so that the discriminator model and the generator model compete with each other to improve the recognition rate.

[0023] An embodiment of the present invention also provides an image segmentation device for evaluating the sub-treatment dose in hepatocellular carcinoma, including:

[0024] An image module for obtaining a medical image dataset of the hepatocellular carcinoma region;

[0025] A network model module for constructing a 3D U-Net generator model and a 3D CNN discriminator model, and coupling the 3D U-Net generator model and the 3D CNN discriminator model into a 3D U-Net GAN network model;

[0026] The 3D U-Net GAN network model is trained using a medical image dataset. The real images in the medical image dataset and the fake images randomly generated by the generator model are used as the inputs of the 3D U-Net GAN network model. The preliminary segmentation results are obtained through forward propagation in the 3D U-Net GAN network model. The losses of the generator model and the discriminator model in the 3D U-Net GAN network model are obtained according to the loss function. The parameters of the 3D U-Net GAN network model are updated through backpropagation in the 3D U-Net GAN network model based on the preliminary segmentation results and the losses of the generator model and the discriminator model. At the same time, the discriminator model is fixed to train the generator model, and the generator model is fixed to train the discriminator model to obtain the trained 3D U-Net GAN network model.

[0027] The segmentation module is used to input the medical image into the trained 3D U-Net GAN network model. The generator model in the trained 3D U-Net GAN network model is used to extract the feature maps of the liver part and the tumor part in the medical image, respectively obtaining the structural feature maps including three-dimensional stereo structures and morphological structures and the radioactive element feature maps including radioactive element distributions and radionuclides of the liver part and the tumor part. The discriminator model is used to screen and remove the fake feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part, and the remaining true feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part are respectively summed up and imaged to obtain the segmentation images of the liver part and the tumor part. The segmentation images of the liver part and the tumor part are joined to obtain the segmented image.

[0028] An embodiment of the present invention also provides an electronic device, including a memory and a processor;

[0029] The memory is used to store a computer program;

[0030] When the processor is used to execute the computer program stored in the memory, the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described above are implemented.

[0031] An embodiment of the present invention also provides a computer-readable storage medium for storing a computer program, and when the computer program is executed by a processor, the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described above are implemented.

[0032] An embodiment of the present invention provides an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma. Compared with the prior art, its beneficial effects are as follows:

[0033] The present invention trains a 3D U-Net GAN network model using a medical image dataset, and alternately trains the generator model and the discriminator model in the 3D U-Net GAN network model using real images and fake images randomly generated by the generator model, so that the trained generator model can generate more accurate and detailed segmentation results for medical images, the trained discriminator model can more accurately evaluate the authenticity of the output of the generator model, and the trained network can accurately reflect the patient's own anatomical structure when segmenting medical images. 10 The distribution of B in the body and the radionuclide kinetics, so as to accurately display the information in the medical image and facilitate application. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic flowchart of medical image preprocessing segmentation based on 3D U-Net GAN for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention;

[0035] Figure 2 It is a schematic flowchart of the Monte Carlo dose software interface method for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention;

[0036] Figure 3 It is a schematic structural diagram of 3D U-Net GAN for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention;

[0037] Figure 4 It is a schematic diagram of the segmentation result of the patient's liver CT image for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention; (a) is the CT image of the patient's liver region; (b) is the extracted patient's liver CT image; (c) is the segmentation result of the patient's liver;

[0038] Figure 5 It is a schematic diagram of the patient's liver voxel model for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention;

[0039] Figure 6 It is a schematic diagram of the patient's liver energy deposition result for an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0040] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings. Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0041] See Figures 1 to 2 , an embodiment of the present invention provides an image segmentation method for evaluating neutron therapy dose in hepatocellular carcinoma, including a medical image segmentation method based on deep learning and a Monte Carlo software interface method.

[0042] Specifically, the medical image segmentation method based on deep learning includes the following steps:

[0043] S1: Read images from the medical image dataset of hepatocellular carcinoma patients and convert them into a specific format for convenient subsequent processing.

[0044] S2: Preprocess the images converted into a specific format, including operations such as denoising and enhancing contrast, to ensure the accuracy and stability of the subsequent segmentation process.

[0045] S3: Perform data augmentation processing on the preprocessed images, including but not limited to operations such as rotation, flipping, and scaling, to increase the diversity of data and improve the generalization ability of the model.

[0046] S4: Build a generator model.

[0047] S5: Build a discriminator model.

[0048] S6: Define a loss function.

[0049] S7: Use the prepared training dataset to train the 3D U-Net GAN model, and optimize the model parameters through iteration to make it better adapt to the segmentation tasks of complex structures such as the liver and tumors.

[0050] S8: Use the trained 3D U-Net GAN model to segment the patient's liver and tumor parts.

[0051] S9: Output the segmented images to provide visual references for doctors to assist them in making more accurate diagnoses and treatment plans.

[0052] Specifically, the Monte Carlo software interface method includes the following steps:

[0053] S1: Establish an individualized voxel model of the patient based on the segmented images.

[0054] S2: Based on medical imaging technology and chemical analysis methods, determine the parameters required for Monte Carlo software calculation, such as the density and elemental composition of tissues and organs.

[0055] S3: Convert the model geometric information, physical information, chemical information, etc. input by the user into the input file form of Monte Carlo software.

[0056] S4: Call Monte Carlo software for dose calculation.

[0057] S5: Read the data provided by the radiation monitoring system and determine the total number of neutron nuclear reactions.

[0058] S6: Display the calculation results of the absorbed dose and draw relevant pictures.

[0059] A specific embodiment of the medical image segmentation method based on deep learning of the present invention is as follows:

[0060] Step S1: Read images from the medical image dataset of hepatocellular carcinoma patients and convert them into a specific format for subsequent processing.

[0061] In step S1, use the PyDicom library in the Python language to read medical images in Dicom format, or use the pynrrd library to read medical images in nnrd format, use the h5py library to read medical images in h5 format, use the PIL library to read medical images in other formats, etc.; use the NumPy library of Python to convert the medical images that have been read into the specific format it provides for subsequent steps.

[0062] Step S2: Preprocess the images converted into a specific format, including operations such as denoising and enhancing contrast, to ensure the accuracy and stability of the subsequent segmentation process.

[0063] In step S2, use the OpenCV library, Scikit-image library, and PIL library for image preprocessing; including applying Gaussian filtering, median filtering, or bilateral filtering techniques to effectively reduce the noise in the image and improve the accuracy of subsequent processing steps; including applying histogram equalization, adaptive histogram equalization, contrast stretching, etc. techniques to enhance the contrast of the image, making different regions in the image easier to distinguish, thus contributing to the accuracy of the segmentation process; including applying geometric transformation operations such as resizing the image and cropping the image to optimize the image data, making it more suitable for subsequent processing and analysis, and improving the performance and accuracy of the model.

[0064] Step S3: Perform data augmentation processing on the preprocessed images, including but not limited to operations such as rotation, flipping, and scaling, to increase the diversity of the data and improve the generalization ability of the model.

[0065] In step S3, the Albumentations library in the Python language is used to perform data augmentation operations on the preprocessed images; including random rotation, horizontal flipping, vertical flipping, random cropping and resizing, and random adjustment of brightness and contrast, etc.; thereby increasing the diversity and robustness of data samples, improving the generalization ability of the model, and alleviating the overfitting phenomenon that may occur during the training process.

[0066] Step S4: Build a generator model.

[0067] In step S4, a 3D U-Net network model is established based on the Pytorch library in the Python language, which adds some convolutional layers and pooling layers between the encoder and the decoder to adapt to 3D data; the generator needs to generate realistic samples from random noise or other inputs, and generates new samples similar to real samples by learning the distribution and features of the data.

[0068] Step S5: Build a discriminator model.

[0069] In step S5, a 3D CNN model is built using the Pytorch library in the Python language to distinguish the authenticity from real samples and generated samples using different network structures; the discriminator needs to be more focused on the classification energy and authenticity judgment of the input samples, rather than just performing feature extraction and image reconstruction; the input of the discriminator is the segmentation result generated by the generator or the real label, and the output is a scalar value (0 represents the generated fake sample, 1 represents the real sample).

[0070] Step S6: Define the loss function.

[0071] In step S6, the loss functions provided in the PyTorch library are called, such as: mean squared error loss function, cross-entropy loss function, binary cross-entropy loss function, weighted binary cross-entropy loss function, etc.; in addition, custom loss functions are also supported for the training of neural networks.

[0072] Step S7: Use the prepared training dataset to train the 3D U-Net GAN model. The structure of the 3D U-Net GAN model is as Figure 3 shown. By iteratively optimizing the model parameters, it can better adapt to the segmentation tasks of complex structures such as the liver and tumors.

[0073] In step S7, the real image and the fake image generated by the generator are used as inputs, and through forward propagation, the generated segmentation result is obtained. The losses of the generator and the discriminator are calculated according to the loss function, and then backpropagation and parameter update are performed. An alternating training strategy is adopted. First, the discriminator is fixed to train the generator, and then the generator is fixed to train the discriminator, so that the two models compete with each other and gradually improve their performance. In this process, whether to use the CUDA acceleration function is selected based on the actual hardware device situation. If it is true, the CUDA acceleration function is used; if it is false, the CUDA acceleration function is not used, but the CPU is directly used for computing.

[0074] Step S8: Use the trained 3D U-Net GAN model to segment the patient's liver and tumor parts.

[0075] In step S8, the preprocessed 3D medical image is passed as an input to the 3D U-Net GAN model. Through the forward propagation process, corresponding segmentation results will be generated according to the feature and distribution information of the input image, and these results can be regarded as predictions of the structural boundaries or regions of interest in the image.

[0076] Step S9: Output the segmented image to provide a visual reference for doctors to assist them in making more accurate diagnoses and treatment plans.

[0077] In step S9, the OpenCV library and PIL library in the Python language are used to output the segmented image in different formats, such as image formats like JPEG, PNG, BMP, TIFF, etc.

[0078] Among them, in the encoder part: the spatial dimension of the image is gradually reduced through convolutional layers, activation functions, and pooling layers, while extracting feature information. For each layer of neurons (starting from the first hidden layer), its input value is calculated, which usually involves the following operations: calculating the weighted sum, adding the bias, and applying the activation function.

[0079] Among them, in the decoder part: the spatial dimension of the image is gradually restored through deconvolutional layers and skip connections, while combining the feature information extracted by the encoder to generate the segmentation result. For each layer of neurons (starting from the first hidden layer), its input value is calculated, which usually involves the following operations: calculating the weighted sum, adding the bias, and applying the activation function.

[0080] Among them, the forward propagation of the generator part: the input data is gradually increased in dimension through a series of neural network layers (usually convolutional layers, activation functions, and upsampling layers) to generate an image. The generator usually consists of multiple layers, including but not limited to: fully connected layers or convolutional layers, activation functions, upsampling layers. In some variants, the generator may include skip connections. Finally, the generator outputs the generated image through an activation function.

[0081] Among them, the forward propagation of the discriminator part: simultaneously receives real images and fake images generated by the generator as inputs, extracts features and classifies them through a series of neural network layers (usually convolutional layers and activation functions), and gradually reduces the spatial dimension of the image through pooling layers or strided convolutions to extract more abstract features. The final output is the discriminator's prediction result on whether the input image is real or fake.

[0082] During the forward propagation process, various regularization techniques may also be applied, such as dropout, weight decay, etc., to prevent overfitting and improve the generalization ability of the model.

[0083] The specific embodiment of the Monte Carlo software interface method of the present invention is as follows:

[0084] Step S1: Establish a patient-specific voxel model based on the segmented image.

[0085] In step S1, a voxel model is established by a single or multiple hexahedrons, as Figure 5 shown, which is used to represent the anatomical structure of a liver cancer patient for subsequent Monte Carlo calculations.

[0086] Step S2: Based on medical imaging techniques and chemical analysis methods, determine the parameters required for Monte Carlo software calculations, such as the density and elemental composition of tissues, organs, and lesion areas.

[0087] In step S2, medical imaging techniques such as: X-rays, CT scans, and MRIs, etc., as Figure 4 shown is the CT image segmentation result of a patient's liver, where (a) is the CT image of the patient's liver area; (b) is the CT image of the patient's liver extracted from the CT image of the patient's liver area; (c) is the segmentation result of the patient's liver; these techniques can provide information about density and elemental composition; chemical analysis methods such as: atomic absorption spectrometry, atomic emission spectrometry, mass spectrometry, metabolomics, and proteomics, etc., usually require collecting tissue samples and performing laboratory analysis, and can determine the elemental composition and content in tissues or organs.

[0088] Step S3: Convert the model geometric information, physical information, chemical information, etc. input by the user into the input file form of the Monte Carlo software.

[0089] In step S3, the geometric information of the model, such as information reflecting the geometric structure like the length, width, height, etc. of the voxels, and spatial information like the distance between the radioactive source and the patient's target area; the physical information of the model, such as information reflecting physical parameters like the particle type, source strength of the radioactive source, beam intensity, emission angle, energy spectrum, and density of tissues and organs; the chemical information of the model includes information reflecting chemical parameters such as the elemental composition of tissues and organs.

[0090] Step S4: Call Monte Carlo software for dose calculation.

[0091] In step S4, call Monte Carlo software for dose calculation, such as MCNP, Geant4, GATE, etc.

[0092] Step S5: Read the data provided by the radiation monitoring system and determine the total number of neutron nuclear reactions.

[0093] In step S5, input the data provided by relevant radiation monitoring instruments such as neutron measuring instruments, activation method detectors, thermoluminescent dosimeters, γ-ray detectors, etc., and then determine the total number of neutron nuclear reactions.

[0094] Step S6: Display the dosimetry calculation results and draw relevant pictures.

[0095] In step S6, use the Matplotlib library in the Python language to draw pictures of the relevant dosimetry calculation results, as Figure 6 shown.

[0096] In this embodiment, in the process of determining parameters required for Monte Carlo software calculation such as the density and elemental composition of tissues, organs, and lesion areas, in addition to medical imaging techniques and chemical analysis methods, it can also be determined through data such as Publication No. 110 of the International Commission on Radiological Protection (ICRP), databases, etc.

[0097] In this embodiment, Monte Carlo software is a general software package for calculating various particle transport problems based on the Monte Carlo method; Monte Carlo software such as MCNP, Geant4, GATE, etc.; use Monte Carlo software to perform particle transport calculations during the treatment process to obtain the average dose value of the lesion area in a single nuclear conversion process of a hepatocellular carcinoma patient, and its calculation formula is as follows:

[0098]

[0099] Where: D represents the average dose value; the average dose value represents the magnitude of the average ionization energy dE absorbed by a unit mass of matter dm under the action of ionizing radiation; this method supports calling Monte Carlo software to calculate the average ionization energy during the particle transport process in the lesion area, and calculates based on the mass of the lesion area obtained from the voxel model, thereby obtaining the average dose value.

[0100] In this embodiment, according to the definition of the Medical Internal Radiation Dose Committee, the absorbed dose of the target area can be expressed as:

[0101]

[0102] Where: D T represents the absorbed dose of the target area; S represents the average absorbed dose of the target area in a single nuclear transformation process (calculated by Monte Carlo software); represents the time-integrated activity.

[0103] In the present invention, for the 3D U-Net GAN model, it combines the idea of the generative adversarial network (GAN), enabling the model to better maintain the authenticity of the image during segmentation; the generator model can generate more accurate and detailed segmentation results, while the discriminator model evaluates the authenticity of the output of the generator model; this adversarial training can prompt the generator model to continuously optimize, thereby improving the accuracy of segmentation; the segmented image can provide accurate anatomical information for subsequent dosimetry calculations.

[0104] The present invention combines Monte Carlo simulation technology and medical image segmentation results, and can quickly and accurately establish a patient-specific dose calculation model; the model information, physical information, and chemical information input by the user are converted into the input file form of Monte Carlo software, facilitating dose calculation; the absorbed dose calculation results are displayed and relevant pictures are drawn, providing an intuitive dose assessment result for doctors and patients.

[0105] The embodiment of the present invention also provides an image segmentation device for evaluating the neutron therapy dose of hepatocellular carcinoma, including:

[0106] An image module for acquiring a medical image dataset of the hepatocellular carcinoma region.

[0107] A network model module for constructing a 3D U-Net generator model and a 3D CNN discriminator model, and coupling the 3D U-Net generator model and the 3D CNN discriminator model into a 3D U-Net GAN network model.

[0108] The 3D U-Net GAN network model is trained using a medical image dataset. The real images in the medical image dataset and the fake images randomly generated by the generator model are used as the inputs of the 3D U-Net GAN network model. The preliminary segmentation results are obtained through forward propagation in the 3D U-Net GAN network model. The losses of the generator model and the discriminator model in the 3D U-Net GAN network model are obtained according to the loss function. According to the preliminary segmentation results and the losses of the generator model and the discriminator model, the parameters of the 3D U-Net GAN network model are updated through backpropagation in the 3D U-Net GAN network model. At the same time, the discriminator model is fixed to train the generator model, and the generator model is fixed to train the discriminator model to obtain the trained 3D U-Net GAN network model.

[0109] A segmentation module is used to input the medical image into the trained 3D U-Net GAN network model. The generator model in the trained 3D U-Net GAN network model is used to extract the feature maps of the liver part and the tumor part in the medical image, respectively obtaining the structural feature maps including three-dimensional stereo structures and morphological structures and the radioactive element feature maps including the distribution of radioactive elements and radionuclides of the liver part and the tumor part. The discriminator model is used to screen and remove the fake feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part, and the remaining true feature elements in the structural feature maps and the radioactive element feature maps of the liver part and the tumor part are respectively summed up and imaged to obtain the segmentation images of the liver part and the tumor part. The segmentation images of the liver part and the tumor part are joined to obtain the segmented image.

[0110] An embodiment of the present invention further provides an electronic device, including a memory and a processor.

[0111] The memory is used to store a computer program.

[0112] When the processor is used to execute the computer program stored in the memory, the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described above are implemented.

[0113] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program. When the computer program is executed by a processor, the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described above are implemented.

[0114] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. An image segmentation method for evaluating the dose of neutron therapy in hepatocellular carcinoma, characterized in that, It includes the following steps: Obtain a medical image dataset of the hepatocellular carcinoma region; Construct a 3D U-Net generator model and a 3D CNN discriminator model, and couple the 3D U-Net generator model and the 3D CNN discriminator model into a 3D U-Net GAN network model; Use the medical image dataset to train the 3D U-Net GAN network model. Take the real images in the medical image dataset and the fake images randomly generated by the generator model as the inputs of the 3D U-Net GAN network model, and obtain a preliminary segmentation result through forward propagation in the 3D U-Net GAN network model. Obtain the losses of the generator model and the discriminator model in the 3D U-Net GAN network model according to the loss function respectively. Update the parameters of the 3D U-Net GAN network model through backpropagation in the 3D U-Net GAN network model according to the preliminary segmentation result and the losses of the generator model and the discriminator model. At the same time, fix the discriminator model to train the generator model, and fix the generator model to train the discriminator model to obtain the trained 3D U-Net GAN network model; Input the medical image into the trained 3D U-Net GAN network model, and use the generator model in the trained 3D U-Net GAN network model to extract the features of the liver part and the tumor part in the medical image, and respectively obtain the structural feature maps including three-dimensional stereoscopic structure and morphological structure and the radioactive element feature maps including radioactive element distribution and radionuclides of the liver part and the tumor part; Use the discriminator model to screen and remove the fake feature elements in the structural feature maps and radioactive element feature maps of the liver part and the tumor part, and respectively sum up the remaining true feature elements in the structural feature maps and radioactive element feature maps of the liver part and the tumor part, and perform image processing to respectively obtain the segmentation images of the liver part and the tumor part, and connect the segmentation images of the liver part and the tumor part to obtain the segmented image.

2. The image segmentation method for evaluating the dose of neutron therapy in hepatocellular carcinoma according to claim 1, wherein, The acquisition of the medical image dataset includes: Obtain the medical images of the hepatocellular carcinoma region; Perform Gaussian filtering technology, median filtering technology, bilateral filtering technology, histogram equalization, adaptive histogram equalization, contrast stretching and geometric transformation processing on the medical images to obtain the medical image dataset of the hepatocellular carcinoma region; Perform random rotation, horizontal flipping, vertical flipping, random cropping, random brightness adjustment or random contrast adjustment processing on the medical image dataset to enhance the data in the medical image dataset.

3. An image segmentation method for evaluating neutron therapy dose in hepatocellular carcinoma according to claim 1, characterized in that, The construction of the 3D U-Net generator model and the 3D CNN discriminator model includes: The 3D U-Net generator model adds multiple convolutional layers and pooling layers between the encoder and the decoder in the model to adapt to 3D data; The 3D U-Net generator model generates realistic samples from random noise or other inputs, and generates new samples similar to the real samples by learning the distribution and features of the sample data; The 3D CNN discriminator model uses different network structures to distinguish authenticity from real samples and newly generated samples similar to real samples.

4. An image segmentation method for evaluating the dose of boron neutron capture therapy in hepatocellular carcinoma according to claim 1, characterized in that, The forward propagation algorithm includes: The forward propagation in the generator part is as follows: The input data is passed through a series of neural network layers to gradually increase the dimension of the input data and generate an image. The generator usually consists of multiple layers, including fully connected layers or convolutional layers, activation functions, and upsampling layers. In some variants, the generator may contain skip connections. Finally, the generator outputs the generated image through an activation function. The forward propagation in the discriminator part is as follows: It receives both real images and fake images generated by the generator as inputs, extracts features and classifies them through a series of neural network layers, and gradually reduces the spatial dimension of the image through pooling layers or strided convolutions to extract more abstract features. The final output is the discriminator's prediction result on whether the input image is real or fake.

5. A method for image segmentation for evaluating the dose of boron neutron capture therapy in hepatocellular carcinoma according to claim 1, characterized in that, The fixed discriminator model is used to train the generator model, and then the fixed generator model is used to train the discriminator model, enabling the discriminator model and the generator model to compete with each other to improve the recognition rate.

6. An image segmentation device for evaluating the dose of neutron therapy in hepatocellular carcinoma, characterized in that, It includes: An image module for obtaining a medical image dataset of hepatocellular carcinoma regions; A network model module for constructing a 3D U-Net generator model and a 3D CNN discriminator model, and coupling the 3D U-Net generator model and the 3D CNN discriminator model into a 3D U-Net GAN network model; Using the medical image dataset to train the 3D U-Net GAN network model, taking the real images in the medical image dataset and the fake images randomly generated by the generator model as the inputs of the 3D U-Net GAN network model, and obtaining a preliminary segmentation result through forward propagation in the 3D U-Net GAN network model; obtaining the losses of the generator model and the discriminator model in the 3D U-Net GAN network model according to the loss function, and updating the parameters of the 3D U-Net GAN network model through backpropagation in the 3D U-Net GAN network model based on the preliminary segmentation result and the losses of the generator model and the discriminator model. At the same time, fix the discriminator model to train the generator model, and fix the generator model to train the discriminator model to obtain the trained 3D U-Net GAN network model; A segmentation module for inputting the medical image into the trained 3D U-Net GAN network model, and using the generator model in the trained 3D U-Net GAN network model to extract the features of the liver part and the tumor part in the medical image, respectively obtaining a structural feature map containing three-dimensional stereo structure and morphological structure and a radioactive element feature map containing radioactive element distribution and radionuclides for the liver part and the tumor part; Using a discriminator model to screen and remove false feature elements in the structural feature maps and radioactive element feature maps of the liver part and the tumor part, and respectively summing up the remaining true feature elements in the structural feature maps and radioactive element feature maps of the liver part and the tumor part, and performing image formation to respectively obtain the segmentation images of the liver part and the tumor part, and connecting the segmentation images of the liver part and the tumor part to obtain the segmented image.

7. An electronic device, characterized in that, Including: A memory and a processor; The memory is used for storing a computer program; When the processor executes the computer program stored in the memory, it implements the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a processor, it implements the steps of an image segmentation method for evaluating the neutron therapy dose in hepatocellular carcinoma as described in any one of claims 1 to 5.

Images