A novel federated semi-supervised learning method and device for lung tumor segmentation

Abstract

The application discloses a novel lung tumor segmentation method and device based on federated semi-supervised learning, and the method comprises the following steps: preprocessing lung tumor data to obtain preprocessed data, and dividing the preprocessed data into a supervised data set and an unsupervised data set; training a 2D U-Net segmentation model based on a federated semi-supervised learning framework by using the preprocessed data to obtain a trained 2D U-Net segmentation model; the training comprises supervised client training and unsupervised client training; inputting lung tumor data to be segmented into the trained 2D U-Net segmentation model, and outputting a lung tumor segmentation result. The application can simultaneously alleviate the problem of data heterogeneity of different clients by using the federated semi-supervised learning framework, so as to reduce the deviation of the model.

Description

Technical Field

[0001] This invention belongs to the technical field of machine learning, specifically relating to a novel method and apparatus for lung tumor segmentation using federated semi-supervised learning. Background Technology

[0002] Federated learning (FL) is an advanced distributed learning concept that utilizes datasets across multiple institutions without any explicit data centralization or sharing. Although initially designed for mobile edge devices, FL has gained increasing attention in the medical field due to its privacy-preserving properties. FL is independent of the type of input data. It is capable of analyzing a wide variety of medical data patterns, from free-text clinical reports to high-dimensional medical images. FL has been used to train predictive models and solve support vector machine problems for analyzing electronic health record (EHR) data. FL has been applied to wearable healthcare using personalized machine learning models. Some researchers have developed an FL framework for privacy-preserving multi-site fMRI classification. Recently, FL has been successfully applied to multi-institutional brain MRI to segment tumors using deep neural networks and improve patient privacy protection.

[0003] Semi-supervised learning leverages the available information from unlabeled data along with the supervision of labeled data to improve the effectiveness and generality of machine learning models. In computer vision, semi-supervised learning has been studied from different perspectives for various applications (e.g., image recognition). To utilize unlabeled data, consistency constraints have been investigated to mitigate the intra- and inter-domain gaps between labeled and unlabeled data. One research trend is the teacher-student model framework, which effectively utilizes consistency constraints between models on labeled and unlabeled data. Another similar framework, "The Noisy Student," achieved state-of-the-art performance on ImageNet (the name of a dataset) classification when jointly trained with a large amount of unlabeled data. Meanwhile, consistency-based model regularization can be achieved through model prediction using unlabeled data, along with data augmentation or preprocessing. Another trend is designing auxiliary supervised tasks for unlabeled data, such as solving jigsaw puzzles or predicting rotation angles. Alternatively, co-training has been applied to semi-supervised image recognition, where different models are trained according to different "views" to learn supplementary information from the data. With the increasing availability of medical datasets, semi-supervised learning techniques for medical image segmentation are also becoming more prevalent.

[0004] Federated semi-supervised learning presents challenges and complexities in utilizing unlabeled data in a distributed learning environment. A literature review reveals very limited research on federated semi-supervised learning for unlabeled medical imaging data. Reference 1 proposes a federated semi-supervised learning framework for pneumonia region segmentation. The basic steps of this method include: 1) converting 3D lung data to the most common resolution of chest CT datasets (0.8mm x 0.8mm x 5.0mm); 2) cropping CT intensity values ​​to the range of -1000 to 0; 3) normalizing the data to the [0,1] interval; 4) cropping the lung data image size to a size suitable for the network model's input; 5) inputting the processed data into the federated semi-supervised learning framework for iterative training. These are the general steps of federated semi-supervised learning for image segmentation. Summary of the Invention

[0005] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a novel method and apparatus for lung tumor segmentation using federated semi-supervised learning. Through the federated semi-supervised learning framework, the potential domain gap between supervised and unsupervised clients can be effectively mitigated.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a novel method and apparatus for lung tumor segmentation using federated semi-supervised learning, comprising the following steps:

[0008] This invention provides a novel method for lung tumor segmentation using federated semi-supervised learning, comprising the following steps:

[0009] The lung tumor data was preprocessed by cropping, resampling and normalization to obtain preprocessed data, and the preprocessed data was divided into supervised dataset and unsupervised dataset.

[0010] The preprocessed data is used to train a 2D U-Net segmentation model based on a federated semi-supervised learning framework to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training.

[0011] The lung tumor data to be segmented is input into the trained 2D U-Net segmentation model, and the lung tumor segmentation result is output.

[0012] As a preferred technical solution, the preprocessing of lung tumor data, including cropping, resampling, and normalization, specifically involves:

[0013] First, the lung tumor data is processed through a segmentation model to segment out the lung area, and then cropped onto the original image to the size of the lung region.

[0014] Secondly, since the resolution of data from each client is different, the resolution of data from each client needs to be resampled to the preset resolution;

[0015] Finally, standardization is performed, taking into account the appropriate window width and window level for the lungs. First, the intensity values ​​of the lung data are truncated to the range of [-250, 400], and then Z-Score is used for standardization to prevent the data from being compressed during normalization.

[0016] As a preferred technical solution, the supervised dataset is divided into a supervised training set, a supervised test set, and a supervised validation set in equal proportions; the unsupervised dataset is divided into an unsupervised training set, an unsupervised test set, and an unsupervised validation set.

[0017] As a preferred technical solution, the specific process of supervised client training is as follows:

[0018] First, the training set of the supervised client generates a predicted label Y1 through the first basic model H(x; θ); then the predicted label Y1 and the label Y are used to calculate the loss.

[0019] Then, validation is performed on a supervised validation set, and each supervised client will generate a validation result P;

[0020] Secondly, the supervised client uploads gradient parameters to the server. Unsupervised client uploads gradient parameters to server

[0021] Server-side aggregation gradient and gradient Then the updated gradient is sent to each supervised client, where the aggregation strategy is:

[0022] Suppose that the result of the validation set during the supervised client's training iteration t is P. i (i∈C s Then, at time t+1, the supervised client gradient update strategy is as shown in the formula:

[0023]

[0024] Among them, C s This refers to a supervised client, C u This refers to an unsupervised client, w s and w u Initial weights for supervised and unsupervised clients respectively. This represents dynamically updated weights. and These represent the gradient parameters for supervised and unsupervised clients, respectively. The formula is:

[0025]

[0026] Finally, the gradient parameters with the best accuracy obtained on the supervised validation set are saved and tested on the supervised test set.

[0027] As a preferred technical solution, the unsupervised client training process is as follows:

[0028] First, the training set of the unsupervised client generates a pseudo-label through the second basic model H(u; θ).

[0029] The training data of the unsupervised client is flipped, rotated, and augmented with Gaussian noise before being passed through a second base model H(u; θ) to generate a predicted label.

[0030] The pseudo-labels generated by the second base model are processed twice from the training set data of the unsupervised client. and prediction labels Apply consistency loss to ensure that predicted values ​​tend to be consistent;

[0031] Assume the gradient generated by the unsupervised client during training iteration t is denoted as . Will After being uploaded to the server, the server aggregates the gradient parameters and returns the global gradient as θ. U ;

[0032] For each unsupervised client, an exponential moving average model EMA_MODEL is defined. The settings for EMA_MODEL are the same as the base model, but the gradient parameter updates for EMA_MODEL differ from those of the second base model. During training iteration t+1, the gradient parameters of EMA_MODEL... The updated formula is:

[0033]

[0034] gradient parameters of EMA_MODEL The update has a global gradient θ U Local gradients of unsupervised clients The weighted summation, where β represents the threshold for the weighted summation;

[0035] Finally, the unsupervised client was tested on the unsupervised test set using the exponential moving average model EMA_MODEL.

[0036] In another aspect, the present invention provides a novel federated semi-supervised learning lung tumor segmentation system, applied to the aforementioned novel federated semi-supervised learning lung tumor segmentation method, including a preprocessing module, a model training module, and a segmentation module;

[0037] The preprocessing module is used to preprocess lung tumor data to obtain preprocessed data, and to divide the preprocessed data into supervised datasets and unsupervised datasets.

[0038] The model training module is used to train the 2DU-Net segmentation model based on the federated semi-supervised learning framework using the preprocessed data to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training.

[0039] The segmentation module is used to input the lung tumor data to be segmented into the trained 2D U-Net segmentation model and output the lung tumor segmentation result.

[0040] In another aspect, the present invention provides an electronic device, characterized in that the electronic device comprises:

[0041] At least one processor; and,

[0042] A memory communicatively connected to the at least one processor; wherein,

[0043] The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the novel federated semi-supervised learning method for lung tumor segmentation.

[0044] In another aspect, the present invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the novel federated semi-supervised learning method for lung tumor segmentation.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] 1. This invention uses a federated semi-supervised learning framework to alleviate the problem of data heterogeneity among different clients, thereby reducing model bias.

[0047] 2. In the supervised client, the present invention dynamically aggregates gradient parameters, effectively utilizing the quality information of the data from each client.

[0048] 3. This invention introduces an exponential moving average model EMA_MODEL into the unsupervised client. The gradient parameters of the previous time step are considered in each iteration of training, so that each unsupervised client can pay more attention to its own dataset. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating a novel federated semi-supervised learning method for lung tumor segmentation according to an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the supervised client training process according to an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of the unsupervised client training process according to an embodiment of the present invention;

[0053] Figure 4 This is a block diagram of a novel federated semi-supervised learning lung tumor segmentation system according to an embodiment of the present invention.

[0054] Figure 5 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0056] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0057] Please see Figure 1One embodiment of this application provides a novel federally supervised semi-supervised learning method for lung tumor segmentation, comprising the following steps:

[0058] S1. Perform preprocessing on lung tumor data, including cropping, resampling, and normalization, to obtain preprocessed data. Then, divide the preprocessed data into supervised datasets and unsupervised datasets.

[0059] The preprocessing process is as follows:

[0060] S11, Cropping; First, the lung tumor data is segmented using a segmentation model to divide the lung into parts, and then cropped from the original image to the size of the lung region;

[0061] S12, Resampling; Since the resolution of data from each client is different, the resolution of data from each client needs to be resampled to 1mm x 1mm x 1mm.

[0062] S13. Standardization: To ensure the appropriate window width and level for lung data, the intensity values ​​of the lung data should first be truncated to the range of [-250, 400], and then standardized using Z-Score to prevent data compression during normalization.

[0063] Furthermore, the supervised dataset is divided proportionally into a supervised training set, a supervised test set, and a supervised validation set; the unsupervised dataset is divided into an unsupervised training set, an unsupervised test set, and an unsupervised validation set.

[0064] S2. The preprocessed data is used to train the 2D U-Net segmentation model based on the federated semi-supervised learning framework to obtain the trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training; 2DU-Net is an end-to-end segmentation network that can effectively segment tumor regions.

[0065] S21, please refer to Figure 2 The specific steps for training the supervised client are as follows:

[0066] S211. First, the training set of the supervised client generates a predicted label Y1 through the first basic model H(x; θ); the predicted label Y1 and the label Y are used to calculate the loss.

[0067] S212. Then perform verification on the supervised verification set. Each supervised client will generate a verification result P.

[0068] S213. Supervised client uploads gradient parameters to server. Unsupervised client uploads gradient parameters to server

[0069] Server-side aggregation gradient and gradient Then the updated gradient is sent to each supervised client, where the aggregation strategy is:

[0070] S214. Assume that the result of the validation set is P when a supervised client performs training on the t-th iteration. i (i∈C s Then, at time t+1, the supervised client gradient update strategy is as shown in the formula:

[0071]

[0072] Among them, C s This refers to a supervised client, C u This refers to an unsupervised client, w s and w u Initial weights for supervised and unsupervised clients respectively. This represents dynamically updated weights. and These represent the gradient parameters for supervised and unsupervised clients, respectively. The formula is:

[0073]

[0074] S215. Finally, save the gradient parameters with the best accuracy obtained on the supervised validation set and test them on the supervised test set.

[0075] S22, please refer to Figure 3 The unsupervised client training process is as follows:

[0076] S221. First, the training set of the unsupervised client generates a pseudo-label through the second basic model H(u; θ).

[0077] S222. The training data of the unsupervised client, after being augmented with data such as flipping, rotating, and Gaussian noise, is then passed through the second basic model H(u; θ) to generate a predicted label.

[0078] S223, Pseudo-labels generated twice from the training set data of the unsupervised client by the second base model. and prediction labels Apply consistency loss to ensure that predicted values ​​tend to be consistent;

[0079] S224. Assume that the gradient generated by the unsupervised client during t iterations of training is denoted as... Will After being uploaded to the server, the server aggregates the gradient parameters and returns the global gradient as θ. U ;

[0080] S225. Define an exponential moving average model EMA_MODEL for each unsupervised client. EMA_MODEL is configured the same as the base model, but the gradient parameter updates of EMA_MODEL differ from the second base model. During the (t+1)th iteration of training, the gradient parameters of the exponential moving average model EMA_MODEL... The updated formula is:

[0081]

[0082] gradient parameters of EMA_MODEL The update has a global gradient θ U Local gradients of unsupervised clients The weighted summation, where β represents the threshold for the weighted summation;

[0083] S226. Finally, the unsupervised client is tested on the unsupervised test set using the exponential moving average model EMA_MODEL.

[0084] S3. Input the lung tumor data to be segmented into the trained 2D U-Net segmentation model and output the lung tumor segmentation result.

[0085] In this invention, the supervised client aggregates gradient parameters by dynamically allocating weights, potentially utilizing the quality of data from each client; the unsupervised client updates gradient parameters by introducing an exponential moving average model EMA_MODEL.

[0086] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0087] Based on the same idea as the novel federated semi-supervised learning lung tumor segmentation method in the above embodiments, the present invention also provides a novel federated semi-supervised learning lung tumor segmentation system, which can be used to execute the above-described novel federated semi-supervised learning lung tumor segmentation method. For ease of explanation, the schematic diagram of an embodiment of the novel federated semi-supervised learning lung tumor segmentation system only shows the parts relevant to the embodiments of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and it may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0088] Please see Figure 4 In another embodiment of this application, a novel federal semi-supervised learning lung tumor segmentation system 100 is provided, which includes a preprocessing module 101, a model training module 102 and a segmentation module 103.

[0089] The preprocessing module 101 is used to preprocess lung tumor data to obtain preprocessed data, and to divide the preprocessed data into supervised datasets and unsupervised datasets.

[0090] The model training module 102 is used to train the 2DU-Net segmentation model based on the federated semi-supervised learning framework using the preprocessed data to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training.

[0091] The segmentation module 103 is used to input the lung tumor data to be segmented into the trained 2D U-Net segmentation model and output the lung tumor segmentation result.

[0092] It should be noted that the novel federated semi-supervised learning lung tumor segmentation system of the present invention corresponds one-to-one with the novel federated semi-supervised learning lung tumor segmentation method of the present invention. The technical features and beneficial effects described in the embodiments of the novel federated semi-supervised learning lung tumor segmentation method described above are applicable to the embodiments of the novel federated semi-supervised learning lung tumor segmentation system. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0093] Furthermore, in the above embodiments of a novel federated semi-supervised learning lung tumor segmentation system, the logical division of each program module is merely an example. In practical applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or software implementation convenience. That is, the internal structure of the novel federated semi-supervised learning lung tumor segmentation system can be divided into different program modules to complete all or part of the functions described above.

[0094] Please see Figure 5 In one embodiment, an electronic device is provided for implementing a novel federated semi-supervised learning method for lung tumor segmentation. The electronic device 200 may include a first processor 201, a first memory 202, and a bus, and may also include a computer program stored in the first memory 202 and executable on the first processor 201, such as a novel federated semi-supervised learning lung tumor segmentation program 203.

[0095] The first memory 202 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as the portable hard drive of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 200. Furthermore, the first memory 202 can include both internal and external storage units of the electronic device 200. The first memory 202 can be used not only to store application software and various types of data installed on the electronic device 200, such as the code of the novel federated semi-supervised learning lung tumor segmentation program 203, but also to temporarily store data that has been output or will be output.

[0096] In some embodiments, the first processor 201 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The first processor 201 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the first memory 202 and calls data stored in the first memory 202 to perform various functions of the electronic device 200 and process data.

[0097] Figure 5 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 5 The structure shown does not constitute a limitation on the electronic device 200, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0098] The novel federated semi-supervised learning lung tumor segmentation program 203 stored in the first memory 202 of the electronic device 200 is a combination of multiple instructions that, when run in the first processor 201, can achieve the following:

[0099] Lung tumor data were preprocessed to obtain preprocessed data, which was then divided into supervised datasets and unsupervised datasets.

[0100] The preprocessed data is used to train a 2D U-Net segmentation model based on a federated semi-supervised learning framework to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training.

[0101] The lung tumor data to be segmented is input into the trained 2D U-Net segmentation model, and the lung tumor segmentation result is output.

[0102] Furthermore, if the modules / units integrated in the electronic device 200 are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0103] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0104] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A novel federated semi-supervised learning method for lung tumor segmentation, characterized in that, Includes the following steps: The lung tumor data was preprocessed by cropping, resampling and normalization to obtain preprocessed data, and the preprocessed data was divided into supervised dataset and unsupervised dataset. The preprocessed data is used to train a 2D U-Net segmentation model based on a federated semi-supervised learning framework to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training. The lung tumor data to be segmented is input into the trained 2D U-Net segmentation model, and the lung tumor segmentation result is output. The specific process of supervised client training is as follows: First, the training set of the supervised client generates a predicted label through the first base model H(x; θ). Predict labels With tags Perform loss calculation; Then, validation is performed on a supervised validation set, and each supervised client will generate a validation result. ; Secondly, the supervised client uploads gradient parameters to the server. The unsupervised client uploads gradient parameters to the server. ; Server-side aggregation gradient and gradient Then the updated gradient is sent to each supervised client, where the aggregation strategy is: Suppose that the supervised client's validation set result is as follows during the t-th iteration of training: Then, at time t+1, the supervised client gradient update strategy is as shown in the formula: in, This indicates that there is a supervised client. This refers to an unsupervised client. and Initial weights for supervised and unsupervised clients respectively. This represents dynamically updated weights. and These represent the gradient parameters for supervised and unsupervised clients, respectively. The formula is: Finally, the gradient parameters with the best accuracy obtained on the supervised validation set are saved and tested on the supervised test set. The unsupervised client training process is as follows: First, the training set of the unsupervised client generates a pseudo-label through the second basic model H(u; θ). ; The training data of the unsupervised client is flipped, rotated, and augmented with Gaussian noise before being passed through a second basic model H(u; θ) to generate a predicted label. ; The pseudo-labels generated by the second base model are processed twice from the training set data of the unsupervised client. and prediction labels Apply consistency loss to ensure that predicted values ​​tend to be consistent; Assume the gradient generated by the unsupervised client during training iteration t is denoted as . ,Will After being uploaded to the server, the server aggregates the gradient parameters and returns the global gradient as follows: ; For each unsupervised client, an exponential moving average model EMA_MODEL is defined. The settings for EMA_MODEL are the same as the base model, but the gradient parameter updates for EMA_MODEL differ from those of the second base model. During training iteration t+1, the gradient parameters of EMA_MODEL... The updated formula is: ， ∈[0,1] gradient parameters of EMA_MODEL The update has a global gradient. Local gradients of unsupervised clients The weighted sum, where This represents the threshold for the weighted summation; Finally, the unsupervised client was tested on the unsupervised test set using the exponential moving average model EMA_MODEL.

2. The novel federated semi-supervised learning method for lung tumor segmentation according to claim 1, characterized in that, The preprocessing of lung tumor data, including cropping, resampling, and normalization, specifically involves: First, the lung tumor data is processed through a segmentation model to segment out the lung area, and then cropped onto the original image to the size of the lung region. Secondly, since the resolution of data from each client is different, the resolution of data from each client needs to be resampled to the preset resolution; Finally, standardization is performed, taking into account the appropriate window width and window level for the lungs. First, the intensity values ​​of the lung data are truncated to the range of [-250, 400], and then Z-Score is used for standardization to prevent the data from being compressed during normalization.

3. The novel federated semi-supervised learning method for lung tumor segmentation according to claim 1, characterized in that, The supervised dataset is divided into a supervised training set, a supervised test set, and a supervised validation set in equal proportions; the unsupervised dataset is divided into an unsupervised training set, an unsupervised test set, and an unsupervised validation set.

4. A novel federated semi-supervised learning system for lung tumor segmentation, characterized in that, A novel federated semi-supervised learning method for lung tumor segmentation, applicable to any one of claims 1-3, includes a preprocessing module, a model training module, and a segmentation module; The preprocessing module is used to preprocess lung tumor data to obtain preprocessed data, and to divide the preprocessed data into supervised datasets and unsupervised datasets. The model training module is used to train the 2DU-Net segmentation model based on the federated semi-supervised learning framework using the preprocessed data to obtain a trained 2D U-Net segmentation model; the training includes supervised client training and unsupervised client training. The segmentation module is used to input the lung tumor data to be segmented into the trained 2D U-Net segmentation model and output the lung tumor segmentation result.

5. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a novel federated semi-supervised learning method for lung tumor segmentation as described in any one of claims 1-3.

6. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements a novel federated semi-supervised learning method for lung tumor segmentation as described in any one of claims 1-3.

Images