CN116109608A - Tumor segmentation method, device, equipment and storage medium - Google Patents
Tumor segmentation method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN116109608A CN116109608A CN202310153706.0A CN202310153706A CN116109608A CN 116109608 A CN116109608 A CN 116109608A CN 202310153706 A CN202310153706 A CN 202310153706A CN 116109608 A CN116109608 A CN 116109608A
- Authority
- CN
- China
- Prior art keywords
- tumor segmentation
- image
- model
- segmentation model
- tumor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/54—Extraction of image or video features relating to texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30056—Liver; Hepatic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Public Health (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Quality & Reliability (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Radiology & Medical Imaging (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Pathology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The application discloses a tumor segmentation method, a device, equipment and a storage medium, which relate to the technical field of medical image processing and comprise the following steps: generating a normalized image and an encoded image based on the original medical radiological image; optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on the combination of local features and global features; and inputting the original medical radiation image, the standardized image and the coded image into a target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the model. According to the method, the model is trained based on the federal learning mode of parameter exchange, and three types of images are used as the input of the model, so that the problem of poor generalization of a tumor segmentation network can be effectively solved, and the segmentation accuracy of tumors in medical radiation images is improved.
Description
Technical Field
The present invention relates to the field of medical image processing technologies, and in particular, to a method, an apparatus, a device, and a storage medium for tumor segmentation.
Background
In recent years, many computer-aided diagnosis systems mainly employ techniques based on computer vision and artificial intelligence to correctly recognize boundary differences between two objects, even in minute shapes. In order to overcome the difficulty in segmenting tumors and livers due to low contrast, irregular shapes and boundary blurring between liver tissues and touching organs, many methods have been developed, and conventional methods generally segment liver tumors by manually setting features, such as gray level, texture, shape and structural features of livers, wherein probability maps, variable models, graph segmentation methods and the like are widely used in the field of segmenting liver tumors, and have good segmentation performance, and conventional methods require manual intervention, so that in the case of some irregular livers, the manual intervention may be better, but also have the disadvantage that the segmentation result of livers is greatly affected by the manually setting features, that is, due to the dependence of the conventional segmentation method on manpower, the final segmentation result may deviate due to the difference of professional competence of operators, for example, the conventional regional growing method, wherein seed points must fall in correct regions to obtain relatively accurate segmentation results, so that doctors often need experience or expert operations to select seed points. The low contrast of each organ of the abdomen CT image (i.e. the transverse plane scanning image) and the organ difference between lesion tissues and individuals bring difficulty to the segmentation of liver tumors, and although the traditional methods for segmenting liver and tumor images are many, the liver segmentation accuracy is greatly improved, but the methods are difficult to meet the requirements of clinical practical application. As another example, a segmentation method based on deep learning, specifically, a liver tumor segmentation technique based on u-net, although a tumor can be effectively determined, the tumor edge segmented by the segmentation technique is still not accurate enough due to the blur of the tumor edge and the unavoidable noise problem existing in the CT image, and a small tumor is easily processed as noise when noise is processed. The methods do not effectively solve the noise problem in CT images, and the segmentation accuracy of tumor edges is not improved, so that the performance in clinical application is not good enough. The liver tumor segmentation method based on deep learning is driven by a large amount of training data, which has strong pertinence to the training data, and currently, the liver tumor segmentation method based on deep learning has the problem of poor generalization in practical use of hospitals, namely, the liver tumor segmentation model obtained by training data of one hospital has good liver tumor segmentation performance in the hospital, but the liver tumor segmentation model in another hospital has unsatisfactory performance, and has some solutions to the problem of poor generalization, for example, a federal learning means is adopted, but a central node, namely, a central hospital, is needed for integrating and distributing all data, and accordingly, the problems of poor safety and data leakage can occur.
In summary, the existing tumor segmentation technology has the problems of artificial dependence, poor generalization, poor safety and easy data leakage, and the current segmentation method based on deep learning cannot effectively solve the noise problem in the CT image. Therefore, how to provide a solution to the above technical problem is a problem that a person skilled in the art needs to solve.
Disclosure of Invention
In view of the above, the present invention aims to provide a tumor segmentation method, device, apparatus and storage medium, which can effectively solve the problem of poor generalization of a tumor segmentation network, protect data privacy, and effectively solve the problem of noise in medical radiological images, thereby improving the segmentation accuracy of tumors in the medical radiological images. The specific scheme is as follows:
in a first aspect, the present application discloses a tumor segmentation method comprising:
generating a corresponding standardized image and an encoded image based on the original medical radiological image;
optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination;
and inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
Optionally, the generating the corresponding standardized image and the coded image based on the original medical radiological image includes:
performing Z-Score standardization processing on the original medical radiation image to obtain a corresponding standardized image;
coding the original medical radiation image according to a preset coding algorithm to obtain a corresponding coded image; the preset encoding algorithm is a local gradient direction algorithm which is predefined based on gradient amplitude and gray scale intensity.
Optionally, the performing the Z-Score normalization on the original medical radiological image to obtain a corresponding normalized image includes:
and performing Z-Score normalization processing on non-zero pixels in the original medical radiation image to obtain a corresponding normalized image.
Optionally, the encoding the original medical radiation image according to a preset encoding algorithm to obtain a corresponding encoded image includes:
and performing edge detection on the original medical radiation image to obtain a corresponding edge detection result, and encoding the original medical radiation image based on a preset encoding algorithm and the edge detection result to generate a corresponding encoded image.
Optionally, the performing edge detection on the original medical radiation image to obtain a corresponding edge detection result includes:
and performing edge detection on the original medical radiation image by using a kirsch operator edge detection algorithm to obtain a corresponding edge detection result.
Optionally, the optimizing training is performed on the first tumor segmentation model which is initially trained in advance by the federal learning mode based on parameter exchange to obtain a trained target tumor segmentation model, which includes:
acquiring a first tumor segmentation model which is subjected to initial training in a local hospital in advance;
carrying out one or more model parameter exchanges on the first tumor segmentation model subjected to initial training in advance and a second tumor segmentation model subjected to initial training in other hospitals selected randomly in each round to obtain a tumor segmentation model subjected to parameter exchange; and carrying out optimization training on the tumor segmentation model subjected to parameter exchange to obtain a trained target tumor segmentation model.
Optionally, in each model parameter exchange process of the first tumor segmentation model and the second tumor segmentation model, the method includes:
randomly selecting a first parameter to be exchanged in the first tumor segmentation model, and randomly selecting a second parameter to be exchanged in the second tumor segmentation model;
exchanging the first parameter to be exchanged in the first tumor segmentation model with the second parameter to be exchanged in the second tumor segmentation model.
In a second aspect, the present application discloses a tumor segmentation apparatus comprising:
an image generation module for generating a corresponding standardized image and an encoded image based on the original medical radiological image;
the model training module is used for carrying out optimization training on a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination;
and the tumor segmentation module is used for inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
In a third aspect, the present application discloses an electronic device comprising:
a memory for storing a computer program;
a processor for executing the computer program to implement the steps of the previously disclosed tumor segmentation method.
In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein the computer program when executed by a processor implements the steps of the previously disclosed tumor segmentation method.
It can be seen that the present application provides a tumor segmentation method comprising: generating a corresponding standardized image and an encoded image based on the original medical radiological image; optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination; and inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model. Therefore, the model is trained by the federal learning mode based on parameter exchange, so that the problem of poor generalization of a tumor segmentation network can be effectively solved, data privacy is protected, three images, namely an original medical radiation image, a standardized image and a coded image, are used as the input of a target tumor segmentation model, the noise problem in the medical radiation image can be effectively solved, and the segmentation precision of tumors in the medical radiation image can be improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for segmenting a tumor disclosed in the present application;
FIG. 2 is a flowchart of a specific tumor segmentation method disclosed in the present application;
FIG. 3 is a schematic view of a medical radiological image Z-source normalization result disclosed in the present application;
FIG. 4 is a schematic diagram of image encoding using a local gradient direction algorithm as disclosed herein;
FIG. 5 is a schematic diagram of an image encoding result disclosed in the present application;
FIG. 6 is a schematic diagram of a framework for extracting liver and tumor boundaries as disclosed herein;
FIG. 7 is a schematic view of a tumor segmentation apparatus according to the disclosure;
fig. 8 is a block diagram of an electronic device disclosed in the present application.
Description of the embodiments
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
At present, the existing tumor segmentation technology has the problems of artificial dependence, poor generalization, poor safety and easy data leakage, and the current segmentation method based on deep learning cannot effectively solve the noise problem in the medical radiation image. Therefore, the application provides a tumor segmentation scheme, which can effectively solve the problem of poor generalization of a tumor segmentation network, protect data privacy and effectively solve the noise problem in medical radiation images, so that the segmentation precision of tumors in the medical radiation images can be improved.
The embodiment of the invention discloses a tumor segmentation method, which is shown in fig. 1 and comprises the following steps:
step S11: corresponding standardized images and encoded images are generated based on the original medical radiological images.
In the embodiment of the present application, the corresponding standardized image and the encoded image are generated based on the original medical radiation image, it can be understood that, because the boundary and the lesion edge of the liver, that is, the tumor edge, in the whole liver have similar intensity values, it is generally challenging to accurately distinguish the boundary and the tumor edge of the liver, the original medical radiation image may be subjected to the related preprocessing operation, so as to obtain two types of medical radiation images, that is, the standardized image and the encoded image, and the original medical radiation image, and the standardized image and the encoded image are used as the input of the tumor segmentation model, so that the segmentation accuracy can be improved, and the original medical radiation image may include, but is not limited to, a CT image or an MRI (Magnetic Resonance Imaging ) image.
Step S12: optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination.
In this embodiment, the first tumor segmentation model that has been initially trained is optimized and trained based on the federal learning method of parameter exchange to obtain the trained target tumor segmentation model. It can be understood that the method using federal learning of parameter exchange as model training means can train a target tumor segmentation model with strong generalization capability on the premise of protecting data privacy.
Specifically, a first tumor segmentation model which is subjected to initial training in a local hospital is obtained; carrying out one or more model parameter exchanges on the first tumor segmentation model subjected to initial training in advance and a second tumor segmentation model subjected to initial training in other hospitals selected randomly in each round to obtain a tumor segmentation model subjected to parameter exchange; and optimizing and training the tumor segmentation model subjected to parameter exchange by using the training set acquired by the local hospital to obtain a trained target tumor segmentation model. In the process of exchanging the parameters of each round of models of the first tumor segmentation model and the second tumor segmentation model, randomly selecting a first parameter to be exchanged in the first tumor segmentation model and randomly selecting a second parameter to be exchanged in the second tumor segmentation model; exchanging the first parameter to be exchanged in the first tumor segmentation model with the second parameter to be exchanged in the second tumor segmentation model, namely exchanging model parameters of a plurality of rounds in each round of training, wherein the exchanged model parameters of each round are randomly selected, so that the exchanged model parameters of each round are different.
For example, each hospital separately trains its own tumor segmentation model, each hospital is equal, after training the tumor segmentation model, each hospital randomly or manually selects another hospital as its own exchange object, and then two hospitals exchange one part of the model, for example, the model has 100 ten thousand parameters, randomly divided into 1 hundred parts, each part has 1 ten thousand parameters, and the two hospitals exchange one part of the 100 parameters, and the one part is randomly transformed.
It should be noted that the model in the federal learning method based on parameter exchange is randomly paired, a central node is not needed, but an organization structure is needed, meanwhile, because each part of exchanged parameters are random and can be transformed each time, the other party is not known to exchange part of model parameters with all other parties when mutually trained, in particular, all local training parties are regarded as central nodes, the initialization action of the exchanged parameters can be initiated by any one data owner, or can be initiated by a third party initialization analyzer which can only send initial model parameters or part of random parameter sets, so that medical data cannot be shared with other institutions, and the data owners cannot share complete parameter sets with other people, but are replaced by random and partial parameter sets, therefore, as a data owner, participator and an analyzer only can obtain partial or random models, the data used for training can be prevented from backtraining the data of the hospital, namely, the hospital can test the hospital and the hospital by obtaining partial data, the data can be further trained by the hospital, the data can be further exchanged by one round of the hospital, namely, the data can be exchanged with other models can be further trained by one round of data can be exchanged, and the model can be further exchanged by a plurality of times, and the data can be further exchanged between the hospital can be saved, each round of data can be exchanged, and the model can be exchanged by the model can be further trained, and the model can be exchanged by the data can be exchanged by a round of another round of data can be exchanged, and trained by the model can be further trained by the data can be completely exchanged by the data and the model can be completely exchanged, and the model can be completely trained, and can be completely exchanged, and completely trained, and can be completely and completely exchanged, and can be completely and completely exchanged, and can be saved, the method can enable information exchange and model exchange between hospitals, so that the model has the capability of strengthening and protecting data privacy. That is, the tumor segmentation neural network is optimally trained by the federal learning means of parameter exchange, so that the problem that the current tumor segmentation networks have good segmentation precision in one hospital but have poor performance in another hospital can be solved, and the privacy of data can be protected.
It can be understood that if only the tumor CT image data of one hospital is used for the tumor segmentation network to perform learning training, the trained tumor segmentation network is likely to perform well only on the one hospital, the performance of using the tumor segmentation network in another hospital is poor, a large number of medical data sets are required for training the tumor segmentation network, but the data sets which can be provided by each hospital are very limited, the generalization capability of the tumor segmentation network can be improved by adopting the method that three types of images are used as data input by combining the dual-path tumor segmentation network with the federal learning of parameter exchange as the network training means, a tumor segmentation model with high precision can be trained under the condition of the limited data sets of each hospital, the problem that the liver tumor segmentation network model of each hospital cannot be trained with high precision due to the limited data sets of the related hospitals can be solved, and the problem that the liver tumor segmentation network model of the other hospitals is poor in practical use of the hospitals can be solved.
Step S13: and inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
It will be appreciated that the use of the original medical radiological image, the normalized image, and the encoded image as inputs to the target tumor segmentation model can improve segmentation accuracy, wherein the additional use of the normalized image and the encoded image as inputs to the model can allow the model to learn faster and with greater accuracy. The target tumor segmentation model is a new network structure based on the combination of local features and global features, and each pixel in the image can be divided into three types, namely, the output of the target tumor segmentation model is a tumor boundary, a liver boundary or other tissues. That is, the target pixels are classified using the additional normalized image and the encoded image as inputs of a model while taking into consideration local features and semi-global features around each target pixel. For example, a 21 x 3patch (i.e., patch) is selected, the center of the current pixel is selected as the local window, and a 64 x 3patch is also selected, the center of the current pixel is selected as the semi-global window, where the number 3 represents three different input images, and the first convolution layer of the network may detect low-level features such as curve points and edges. The semi-global patch can provide more information about similar contacted tissue, so that a segmentation line is drawn between them, resulting in corresponding segmentation results, and the results of the tumor segmentation algorithm are highly dependent on the information extracted from the semi-global window.
In this embodiment, the target tumor segmentation model can help a doctor to quickly and accurately locate a tumor region by segmenting the boundary of the liver tumor more accurately, and accurately segment the boundary of the tumor to help the doctor to perform subsequent treatment, such as accurately cutting the tumor, if the edge position segmentation of the tumor is inaccurate, the boundary of the tumor has great influence on the doctor and the patient, and extra positions except the tumor are removed to cause irreparable damage to the body of the patient, so that the trust of the patient to the doctor is lost. Or the doctor does not completely resect the tumor (malignancy), there is an influence that the cancer cells continue to spread and deteriorate.
Therefore, in the embodiment of the application, the problem of poor generalization of the tumor segmentation network can be effectively solved by training the model in a federal learning mode based on parameter exchange, data privacy is protected, and three types of images, namely an original medical radiation image, a standardized image and a coded image, are used as the input of a target tumor segmentation model, so that the noise problem in the medical radiation image can be effectively solved, and the segmentation precision of tumors in the medical radiation image can be improved.
Referring to fig. 2, an embodiment of the present invention discloses a specific tumor segmentation method, and compared with the previous embodiment, the present embodiment further describes and optimizes the technical scheme.
Step S21: and performing Z-Score normalization processing on the original medical radiation image to obtain a corresponding normalized image.
It will be appreciated that noise in the original medical radiological image is primarily derived from random noise, which can be reduced during imaging by increasing the number of photons, but there is still a significant amount of noise in the medical radiological image received at any clinic or hospital. Therefore, the Z-Score standardization processing is carried out on the original medical radiation image, the noise problem in the original medical radiation image can be effectively solved, the distinguishing degree of contact organs is enhanced, and when the Z-Score standardization processing is carried out on the original medical radiation image, the Z-Score standardization processing is mainly carried out on non-zero pixels in the original medical radiation image, namely, the non-zero pixels in the original medical radiation image are subjected to the Z-Score standardization processing to obtain corresponding standardized images.
Specifically, the mean and unit variance of all non-zero pixels in the original medical radiological image are made zero by means of Z-Score normalization, and the correlation formula is as follows:
Z=(x-μ)/σ；
where σ and μ represent the standard deviation and average value of the non-zero pixel intensities, respectively, and x represents the pixel intensity.
For example, as shown in fig. 3, the first row represents the original medical radiological image, the second and third rows represent the Z-Score normalization output using only non-zero pixels and all pixels, respectively, in each column the original medical radiological image and the image normalization results using different initializations, which pixels within the image should participate in the normalization process, e.g. if all pixels within the image are used, the output image is smoother, but it is very difficult to determine the boundaries of the contacting organ, and gaps between organs can be obtained using only non-zero pixels, thus making identification of organ boundaries easier. The critical area is represented in said fig. 3 using rectangular boxes, it can be seen that the difference between z-score normalization using non-zero pixels and z-score normalization using all pixels, i.e. the white oval object, i.e. the bump created in the rib, is easier to detect and many complex boundaries to identify than the third row in said fig. 3, the second row in said fig. 3.
Step S22: coding the original medical radiation image according to a preset coding algorithm to obtain a corresponding coded image; the preset encoding algorithm is a local gradient direction algorithm which is predefined based on gradient amplitude and gray scale intensity.
It is noted that texture analysis is used for the characterization of a surface, e.g. color, contrast and shape. In the encoding of textures, the image is converted into a new representation according to a predefined encoding algorithm, which is a new encoding technique based on gradient magnitude and gray scale intensity, i.e. LDOG (The Local Direction of Gradient, local gradient direction) encoding algorithm, using local descriptors to create a very useful illumination-invariant representation for texture analysis. The use of a representation of local gradient directions to generate an illumination-invariant representation of the original medical radiological image enables the local representation of the texture to be calculated and based on two adjacent clusters around each pixel selected, wherein segmentation accuracy can be increased, especially at tumor boundaries, due to the recognition capability of the local gradient direction algorithm encoding the texture and insensitivity to noise.
In this embodiment, the encoding method includes encoding the original medical radiological image according to a preset encoding algorithm to obtain a corresponding encoded image, specifically, performing edge detection on the original medical radiological image to obtain a corresponding edge detection result, and encoding the original medical radiological image based on the preset encoding algorithm and the edge detection result to generate a corresponding encoded image, where the edge detection algorithm for performing edge detection on the original medical radiological image may include, but is not limited to, a kirsch operator edge detection algorithm, that is, performing edge detection on the original medical radiological image by using the kirsch operator edge detection algorithm to obtain a corresponding edge detection result, so as to improve the ductility of boundary detection and facilitate extraction of key local shape details.
For example, as shown in fig. 4, in the first step of the encoded image, all significant edges are obtained by using Kirsch filter, the edge direction is mainly characterized by a maximum value generated by one of the nonlinear edge detector filters rotated 45 by one of the four directions, in the second step of the encoded image, a 5×5 window needs to be selected from the filtered image, and the position of the center of the window represents the position of the calculated value in the encoded image, that is, the value calculated in the position of the center of the window represents the encoded value in this position in the encoded image, and all 25 pixels in the window are divided into 4 patches of 2×2 and 4 patches of 1×2 different, in the third step of the encoded image, all values in each patch are sorted in descending order, the maximum value inside the patch remains unchanged, gradient values between their orders are calculated by subtracting adjacent smaller values, and the result of the subtraction is taken as the gradient value, and in the fourth step of the encoded image, all patches are calculated and sorted in ascending order by 8. In a fifth step of encoding the image, all 8 ordered averages are embedded in a 3 x 3 template patch according to their original positions within the 5 x 5 window, wherein each value in the 3 x 3 template patch is marked and replaced with an ordered number, then a positive sign is applied to the N, S, W and E directions, the other directions get a negative sign, finally all averages with their own signs are added together and a final value is generated, which represents the encoded value at the same position in the original image.
For example, as shown in fig. 5, a nonlinear edge detector is applied to a CT image to obtain a correlated encoded result, and in each column, the original image and the result of edge extraction and encoding of the image are displayed, i.e., the first row in fig. 5 represents the original image, the second row in fig. 5 depicts edge detection using eight-direction Kirsch filters, and the third row in fig. 5 represents the result of local gradient direction algorithm encoding. By applying a local gradient direction algorithm for encoding, more information about the image structure, such as texture information, can be extracted.
Step S23: optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination.
Step S24: and inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
For the specific content of the above steps S23 to S24, reference may be made to the corresponding content disclosed in the foregoing embodiment, and no detailed description is given here.
Therefore, in the embodiment of the application, the problem of poor generalization of the tumor segmentation network can be effectively solved by training the model in a federal learning mode based on parameter exchange, data privacy is protected, and three types of images, namely an original medical radiation image, a standardized image and a coded image, are used as the input of a target tumor segmentation model, so that the noise problem in the medical radiation image can be effectively solved, and the segmentation precision of tumors in the medical radiation image can be improved.
For example, as shown in fig. 6, the original CT image is subjected to Z-Score normalization to obtain a corresponding normalized image, the original CT image is subjected to edge detection by using eight direction kirsch filters to obtain a corresponding edge detection result, the original CT image is encoded based on an LDOG encoding algorithm and the edge detection result to generate a corresponding encoded image, and then the original CT image, the normalized image and the encoded image are input to a target tumor segmentation model to obtain a liver segmentation result and a liver tumor segmentation result output by the target tumor segmentation model.
Correspondingly, the embodiment of the application also discloses a tumor segmentation device, referring to fig. 7, the device comprises:
an image generation module 11 for generating a corresponding standardized image and an encoded image based on the original medical radiological image;
the model training module 12 is configured to perform optimization training on a first tumor segmentation model that is initially trained based on a federal learning manner of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination;
and the tumor segmentation module 13 is used for inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result output by the target tumor segmentation model.
From the above, in the embodiment of the present application, the model is trained by the federal learning method based on parameter exchange, which can effectively solve the problem of poor generalization of the tumor segmentation network, protect data privacy, and use three types of images, namely, the original medical radiation image, the standardized image and the encoded image, as the input of the target tumor segmentation model, so that the noise problem in the medical radiation image can be effectively solved, and the segmentation precision of tumors in the medical radiation image can be improved.
In some specific embodiments, the image generating module 11 may specifically include:
the first image generation unit is used for performing Z-Score standardization processing on the original medical radiation image to obtain a corresponding standardized image;
the second image generation unit is used for encoding the original medical radiation image according to a preset encoding algorithm to obtain a corresponding encoded image; the preset encoding algorithm is a local gradient direction algorithm which is predefined based on gradient amplitude and gray scale intensity.
In some specific embodiments, the first image generating module may specifically include:
and the first image generation subunit is used for performing Z-Score normalization processing on non-zero pixels in the original medical radiation image to obtain a corresponding normalized image.
In some specific embodiments, the first image generating module may specifically include:
the edge detection unit is used for carrying out edge detection on the original medical radiation image to obtain a corresponding edge detection result, and encoding the original medical radiation image based on a preset encoding algorithm and the edge detection result to generate a corresponding encoded image.
In some specific embodiments, the edge detection unit may specifically include:
and the edge detection subunit is used for carrying out edge detection on the original medical radiation image by utilizing a kirsch operator edge detection algorithm to obtain a corresponding edge detection result.
In some specific embodiments, the model training module 12 may specifically include:
the model selecting unit is used for acquiring a first tumor segmentation model which is subjected to initial training in the local hospital in advance;
the parameter exchange unit is used for carrying out one or more times of model parameter exchange on the first tumor segmentation model which is subjected to initial training in advance and the second tumor segmentation model which is subjected to initial training in other hospitals selected randomly in each round to obtain a tumor segmentation model after parameter exchange;
and the model training unit is used for carrying out optimization training on the tumor segmentation model subjected to parameter exchange to obtain a trained target tumor segmentation model.
In some specific embodiments, the parameter exchange unit may specifically include:
a parameter to be exchanged subunit, configured to select a first parameter to be exchanged in the first tumor segmentation model randomly, and select a second parameter to be exchanged in the second tumor segmentation model randomly;
and the parameter exchange subunit is used for exchanging the first parameter to be exchanged in the first tumor segmentation model with the corresponding second parameter to be exchanged in the second tumor segmentation model.
Further, the embodiment of the application also provides electronic equipment. Fig. 8 is a block diagram of an electronic device 20, according to an exemplary embodiment, and the contents of the diagram should not be construed as limiting the scope of use of the present application in any way.
Fig. 8 is a schematic structural diagram of an electronic device 20 according to an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input output interface 25, and a communication bus 26. Wherein the memory 22 is used for storing a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the tumor segmentation method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in the present embodiment may be specifically an electronic computer.
In this embodiment, the power supply 23 is configured to provide an operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and an external device, and the communication protocol to be followed is any communication protocol applicable to the technical solution of the present application, which is not specifically limited herein; the input/output interface 25 is used for acquiring external input data or outputting external output data, and the specific interface type thereof may be selected according to the specific application requirement, which is not limited herein.
The memory 22 may be a carrier for storing resources, such as a read-only memory, a random access memory, a magnetic disk, or an optical disk, and the resources stored thereon may include an operating system 221, a computer program 222, and the like, and the storage may be temporary storage or permanent storage.
The operating system 221 is used for managing and controlling various hardware devices on the electronic device 20 and computer programs 222, which may be Windows Server, netware, unix, linux, etc. The computer program 222 may further comprise a computer program capable of performing other specific tasks in addition to the computer program capable of performing the tumor segmentation method performed by the electronic device 20 as disclosed in any of the previous embodiments.
Further, the embodiment of the application also discloses a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is loaded and executed by a processor, the steps of the tumor segmentation method disclosed in any embodiment are realized.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, so that the same or similar parts between the embodiments are referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above details of a tumor segmentation method, device, apparatus and storage medium provided by the present invention, and specific examples are applied herein to illustrate the principles and embodiments of the present invention, and the above examples are only for helping to understand the method and core ideas of the present invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.
Claims (10)
1. A method of tumor segmentation, comprising:
generating a corresponding standardized image and an encoded image based on the original medical radiological image;
optimizing and training a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination;
and inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
2. The tumor segmentation method according to claim 1, wherein the generating respective normalized and encoded images based on the original medical radiological image comprises:
performing Z-Score standardization processing on the original medical radiation image to obtain a corresponding standardized image;
coding the original medical radiation image according to a preset coding algorithm to obtain a corresponding coded image; the preset encoding algorithm is a local gradient direction algorithm which is predefined based on gradient amplitude and gray scale intensity.
3. The tumor segmentation method according to claim 2, wherein the performing the Z-Score normalization on the raw medical radiological image to obtain a corresponding normalized image includes:
and performing Z-Score normalization processing on non-zero pixels in the original medical radiation image to obtain a corresponding normalized image.
4. The tumor segmentation method according to claim 2, wherein the encoding the original medical radiological image according to a preset encoding algorithm to obtain a corresponding encoded image includes:
and performing edge detection on the original medical radiation image to obtain a corresponding edge detection result, and encoding the original medical radiation image based on a preset encoding algorithm and the edge detection result to generate a corresponding encoded image.
5. The tumor segmentation method according to claim 4, wherein the performing edge detection on the original medical radiological image to obtain a corresponding edge detection result includes:
and performing edge detection on the original medical radiation image by using a kirsch operator edge detection algorithm to obtain a corresponding edge detection result.
6. The tumor segmentation method according to any one of claims 1 to 5, wherein the optimizing training of the first tumor segmentation model, which is initially trained, by the federal learning method based on parameter exchange to obtain the trained target tumor segmentation model comprises:
acquiring a first tumor segmentation model which is subjected to initial training in a local hospital in advance;
carrying out one or more model parameter exchanges on the first tumor segmentation model subjected to initial training in advance and a second tumor segmentation model subjected to initial training in other hospitals selected randomly in each round to obtain a tumor segmentation model subjected to parameter exchange;
and carrying out optimization training on the tumor segmentation model subjected to parameter exchange to obtain a trained target tumor segmentation model.
7. The tumor segmentation method according to claim 6, characterized in that during each model parameter exchange of the first tumor segmentation model and the second tumor segmentation model, it comprises:
randomly selecting a first parameter to be exchanged in the first tumor segmentation model, and randomly selecting a second parameter to be exchanged in the second tumor segmentation model;
exchanging the first parameter to be exchanged in the first tumor segmentation model with the second parameter to be exchanged in the second tumor segmentation model.
8. A tumor segmentation apparatus, comprising:
an image generation module for generating a corresponding standardized image and an encoded image based on the original medical radiological image;
the model training module is used for carrying out optimization training on a first tumor segmentation model which is subjected to initial training in advance based on a federal learning mode of parameter exchange to obtain a trained target tumor segmentation model; the first tumor segmentation model is a model obtained by utilizing a training set acquired by a local hospital to perform initial training on a dual-path convolutional neural network constructed based on local feature and global feature combination;
and the tumor segmentation module is used for inputting the original medical radiation image, the standardized image and the coded image into the target tumor segmentation model to obtain a liver segmentation result and a tumor segmentation result which are output by the target tumor segmentation model.
9. An electronic device, comprising:
a memory for storing a computer program;
a processor for executing the computer program to implement the steps of the tumor segmentation method as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program; wherein the computer program when executed by a processor implements the steps of the tumor segmentation method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310153706.0A CN116109608A (en) | 2023-02-23 | 2023-02-23 | Tumor segmentation method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310153706.0A CN116109608A (en) | 2023-02-23 | 2023-02-23 | Tumor segmentation method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116109608A true CN116109608A (en) | 2023-05-12 |
Family
ID=86265438
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310153706.0A Pending CN116109608A (en) | 2023-02-23 | 2023-02-23 | Tumor segmentation method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116109608A (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872325A (en) * | 2019-01-17 | 2019-06-11 | 东北大学 | Full-automatic liver neoplasm dividing method based on two-way Three dimensional convolution neural network |
| US10482603B1 (en) * | 2019-06-25 | 2019-11-19 | Artificial Intelligence, Ltd. | Medical image segmentation using an integrated edge guidance module and object segmentation network |
| CN111600707A (en) * | 2020-05-15 | 2020-08-28 | 华南师范大学 | Decentralized federal machine learning method under privacy protection |
| CN112085743A (en) * | 2020-09-04 | 2020-12-15 | 厦门大学 | Image segmentation method for renal tumor |
| CN113467928A (en) * | 2021-05-20 | 2021-10-01 | 杭州趣链科技有限公司 | Block chain decentralization-based federated learning member reasoning attack defense method and device |
| WO2021204040A1 (en) * | 2020-10-29 | 2021-10-14 | 平安科技（深圳）有限公司 | Federated learning data processing method and apparatus, and device and storage medium |
| CN113806768A (en) * | 2021-08-23 | 2021-12-17 | 北京理工大学 | Lightweight federated learning privacy protection method based on decentralized security aggregation |
| CN113935469A (en) * | 2021-10-26 | 2022-01-14 | 城云科技（中国）有限公司 | Model training method based on decentralized federal learning |
| CN114140478A (en) * | 2022-01-30 | 2022-03-04 | 电子科技大学 | Federal learning method, system, device and medium for medical image segmentation |
| CN114186671A (en) * | 2021-12-07 | 2022-03-15 | 南京大学 | Large-batch decentralized distributed image classifier training method and system |
| CN114519306A (en) * | 2022-02-21 | 2022-05-20 | 山西大学 | Decentralized terminal node network model training method and system |
| CN115170857A (en) * | 2022-05-12 | 2022-10-11 | 杭州电子科技大学 | Pancreatic cancer image identification method based on federal transfer learning |
-
2023
- 2023-02-23 CN CN202310153706.0A patent/CN116109608A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872325A (en) * | 2019-01-17 | 2019-06-11 | 东北大学 | Full-automatic liver neoplasm dividing method based on two-way Three dimensional convolution neural network |
| US10482603B1 (en) * | 2019-06-25 | 2019-11-19 | Artificial Intelligence, Ltd. | Medical image segmentation using an integrated edge guidance module and object segmentation network |
| CN111600707A (en) * | 2020-05-15 | 2020-08-28 | 华南师范大学 | Decentralized federal machine learning method under privacy protection |
| CN112085743A (en) * | 2020-09-04 | 2020-12-15 | 厦门大学 | Image segmentation method for renal tumor |
| WO2021204040A1 (en) * | 2020-10-29 | 2021-10-14 | 平安科技（深圳）有限公司 | Federated learning data processing method and apparatus, and device and storage medium |
| CN113467928A (en) * | 2021-05-20 | 2021-10-01 | 杭州趣链科技有限公司 | Block chain decentralization-based federated learning member reasoning attack defense method and device |
| CN113806768A (en) * | 2021-08-23 | 2021-12-17 | 北京理工大学 | Lightweight federated learning privacy protection method based on decentralized security aggregation |
| CN113935469A (en) * | 2021-10-26 | 2022-01-14 | 城云科技（中国）有限公司 | Model training method based on decentralized federal learning |
| CN114186671A (en) * | 2021-12-07 | 2022-03-15 | 南京大学 | Large-batch decentralized distributed image classifier training method and system |
| CN114140478A (en) * | 2022-01-30 | 2022-03-04 | 电子科技大学 | Federal learning method, system, device and medium for medical image segmentation |
| CN114519306A (en) * | 2022-02-21 | 2022-05-20 | 山西大学 | Decentralized terminal node network model training method and system |
| CN115170857A (en) * | 2022-05-12 | 2022-10-11 | 杭州电子科技大学 | Pancreatic cancer image identification method based on federal transfer learning |
Non-Patent Citations (1)
| Title |
|---|
| AMIRHOSSEIN AGHAMOHAMMADI等: "TPCNN: Two-path convolutional neural network for tumor and 1 liver segmentation in CT images using a novel encoding approach", 《EXPERT SYSTEMS WITH APPLICATIONS》, pages 1 - 39 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Heo et al. | Artificial intelligence in oral and maxillofacial radiology: what is currently possible? | |
| US11593943B2 (en) | RECIST assessment of tumour progression | |
| van Ginneken | Fifty years of computer analysis in chest imaging: rule-based, machine learning, deep learning | |
| Saad et al. | Image segmentation for lung region in chest X-ray images using edge detection and morphology | |
| US20200035350A1 (en) | Method and apparatus for processing histological image captured by medical imaging device | |
| CN110570426B (en) | Image co-registration and segmentation using deep learning | |
| CN106887039A (en) | The three-D imaging method and system of a kind of organ and focus based on medical image | |
| ShanmugaPriya et al. | Efficient fuzzy c-means based multilevel image segmentation for brain tumor detection in MR images | |
| CN106898044B (en) | Organ splitting and operating method and system based on medical images and by utilizing VR technology | |
| Peng et al. | A region‐appearance‐based adaptive variational model for 3D liver segmentation | |
| Sufyan et al. | A novel multimodality anatomical image fusion method based on contrast and structure extraction | |
| Zhou et al. | Multiple abdominal organ segmentation: an atlas-based fuzzy connectedness approach | |
| CN114332132A (en) | Image segmentation method and device and computer equipment | |
| Kaur et al. | Optimized multi threshold brain tumor image segmentation using two dimensional minimum cross entropy based on co-occurrence matrix | |
| CN106780728A (en) | A kind of single organ method for splitting and system based on medical image | |
| CN111161268B (en) | Image processing method, device, electronic equipment and computer storage medium | |
| Veluchamy et al. | Brightness preserving optimized weighted bi‐histogram equalization algorithm and its application to MR brain image segmentation | |
| Viji et al. | Performance evaluation of standard image segmentation methods and clustering algorithms for segmentation of MRI brain tumor images | |
| Sumithra et al. | Modified Global Flower Pollination Algorithm‐based image fusion for medical diagnosis using computed tomography and magnetic resonance imaging | |
| CN116309647B (en) | Method for constructing craniocerebral lesion image segmentation model, image segmentation method and device | |
| Ashour et al. | Advanced optimization-based neutrosophic sets for medical image denoising | |
| CN116433976A (en) | Image processing method, device, equipment and storage medium | |
| CN116109608A (en) | Tumor segmentation method, device, equipment and storage medium | |
| CN113450893B (en) | Brain functional region positioning and side fixing method, device, equipment and storage medium | |
| Rao et al. | Deep learning-based medical image fusion using integrated joint slope analysis with probabilistic parametric steered image filter |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |