Gastric adenocarcinoma pathological image classification system based on federal learning
By adopting federated learning methods in the pathological image classification system of gastric adenocarcinoma, the problem of single data set and poor generalization of classification models is solved, and higher classification accuracy and data security are achieved, which is suitable for data sharing and model updates in multiple medical institutions.
Patent Information
- Application Number
- CN202510110781.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2025-06-27
AI Technical Summary
The prior art is difficult to effectively solve the problem of single sample data sets and poor generalization of classification models in the classification of gastric adenocarcinoma pathological image, resulting in low classification accuracy and difficulty in sharing medical data across regions and across data sets.
The commonality and universality of the classification model are improved through the federated learning update iteration process between the central server and the local server. The system uses trusted third parties to generate and split keys to ensure the security and privacy of medical data, and realize data sharing and model updates in multiple medical institutions.
Through federal learning, the accuracy and robustness of gastric adenocarcinoma pathological image classification is improved, the problem of single sample data set is solved, the safety and privacy of medical data is ensured, and the workload of doctors is reduced.
Smart Images
Figure CN120219787A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of image classification, and particularly to a gastric adenocarcinoma pathological image classification system based on federated learning. Background Art
[0002] The statements in this part merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.
[0003] Gastric adenocarcinoma is the most common type of gastric cancer, accounting for about 90%-95% of all gastric cancers. The most commonly used system for staging gastric cancer is the TNM staging system. Doctors use the results of examinations and imaging scans, etc. to answer three key information parts in the TNM staging to judge the comprehensive metastasis situation of gastric adenocarcinoma in tissue, lymphatic system and blood diffusion, and classify gastric adenocarcinoma into stage 0, stage IA, stage IB, stage IIA, stage IIB, stage IIIA, stage IIIB, stage IIIC and stage IV. Pathological TNM staging is one of the important factors determining subsequent treatment plans and predicting the prognosis of patients. Therefore, the classification of gastric adenocarcinoma pathological images is of great significance. However, the total number of registered pathologists is currently not large, and it takes 5 to 10 years to train an experienced pathologist. The huge gap of pathologists needs to be solved urgently. Therefore, promoting AI-assisted diagnosis is an important breakthrough to change the current situation and improve the level of pathological diagnosis in China.
[0004] In actual hospital applications, the background of gastric adenocarcinoma pathological images is complex, there are many types of targets to be classified, and the features that each target needs to meet are complex. Therefore, the classification model is also very complex. At the same time, if the classification model is directly deployed locally in the hospital, a large amount of computing power is required to meet the requirements of model operation. At the same time, the gastric adenocarcinoma pathological images stored in each medical institution are limited, and there are significant differences in the types of gastric cancer of patients in different regions. The systems trained with data from a single medical institution often have limitations and it is difficult to achieve large-scale, cross-time, cross-space and cross-dataset shared applications of health care data while protecting privacy. Summary of the Invention
[0005] In order to solve the above problems, the present disclosure proposes a gastric adenocarcinoma pathological image classification system based on federated learning, which uses federated learning to effectively improve the generalization and universality of the classification model while ensuring the security of medical data and privacy, solves the problem of single sample data set in traditional training, and improves the accuracy of pathological image classification.
[0006] According to some embodiments, the present disclosure adopts the following technical solutions:
[0007] A gastric adenocarcinoma pathological image classification system based on federated learning includes a central server, a plurality of local servers connected to the central server, and a trusted third party;
[0008] The local server is used to obtain gastric adenocarcinoma pathological images and classify the pathological images based on the classification model; the central server and multiple local servers update the classification model through federated learning; the trusted third party is used to generate and split the key;
[0009] Among them, at the beginning stage of federated learning, the central server distributes global model information to local servers. The local servers use the keys of trusted third parties to decode the model information and train local models on the local servers. After a sufficient number of iterations, the local models are sent to the central server. After receiving the local models sent by all local servers, the central server restores the key, decrypts the local models of the local servers and aggregates them to obtain a new global model and send it to each local server as the classification model of the local server.
[0010] Furthermore, the specific steps of the federated learning include: the central server initializes various parameters, wherein the various parameters include the total number of participating nodes, the number of participating nodes in each round, the number of global iterations, the number of local iterations, the initialization model, and the training function;
[0011] The trusted third party generates a master key, encrypts the ciphertext using the master key, and sends the ciphertext to each local server.
[0012] Furthermore, the local server receives the global model information sent by the central server and saves it as a local model, and trains the local model using the master key and ciphertext received from a trusted third party; after the local model training is completed, the local server uses the private key generated by the master key to encrypt the model parameters, and transmits the encryption result and the private key back to the central server; the central server selects multiple nodes from a number of nodes as participating nodes in this round, receives the encryption result and private key sent, and decrypts them; the central server aggregates the global model parameters based on all the received results of local training based on the local data set, and sends the aggregated global model to all local servers, and this round of global training ends.
[0013] Furthermore, the specific method of using local data sets to optimize the global model is: based on the weights of the global model, use multi-scale image input for training images, and then perform multi-scale feature fusion on the multi-scale input images. By training local data, the global model is optimized using stochastic gradient descent and minimizing the loss function.
[0014] Furthermore, the multi-scale image input uses a Gaussian pyramid, and uses convolution and downsampling operations to obtain a higher layer of image. The multi-scale feature fusion is achieved through a parallel multi-branch network using convolution kernels of different sizes.
[0015] Further, the aggregation formula for aggregating to obtain the global model is:
[0016]
[0017] where M is the number of nodes participating in weight update in this round, is the local model parameter sent by the local server i during the (t + 1)-th global model update, is the updated global model.
[0018] Further, each local server is connected to a number of workstations, and the workstations are used to upload gastric adenocarcinoma pathological images collected by different ultrasonic devices to the local server.
[0019] Further, the workstations are connected to different scanners, and the scanners are used to scan and collect gastric adenocarcinoma pathological images in different ultrasonic devices into the workstations.
[0020] According to some embodiments, the present disclosure adopts the following technical solutions:
[0021] An electronic device includes: a memory for non-temporarily storing computer-readable instructions; and a processor for executing the computer-readable instructions, wherein when the computer-readable instructions are run by the processor, the following steps are performed: obtaining gastric adenocarcinoma pathological images; classifying the gastric adenocarcinoma pathological images by using a classification model updated through federated learning.
[0022] According to some embodiments, the present disclosure adopts the following technical solutions:
[0023] A storage medium includes non-temporarily storing computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the following steps are performed: obtaining gastric adenocarcinoma pathological images; classifying the gastric adenocarcinoma pathological images by using a classification model updated through federated learning.
[0024] Compared with the prior art, the beneficial effects of the present disclosure are:
[0025] A gastric adenocarcinoma pathological image classification system based on federated learning proposed by the present disclosure constitutes an update and iteration process of federated learning between the local server and the central server. This process is continuously repeated. After multiple iterations, the optimal model is finally reached on the central server and then sent to each local server, that is, each medical institution. In this way, each medical institution has the latest model with high accuracy and good robustness. At the same time, the local server combines local private data for training to obtain a classification sub-model suitable for the local area.
[0026] A gastric adenocarcinoma pathological image classification system based on federated learning proposed by the present disclosure can more effectively improve the generalization and universality of the classification model while ensuring the security and privacy of medical data held by each hospital, solve the problem of single sample data set in traditional training, improve the accuracy of gastric adenocarcinoma pathological image classification, and reduce the workload of doctors. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] The accompanying drawings forming a part of this disclosure are used to provide a further understanding of the disclosure. The illustrative embodiments and descriptions thereof of the disclosure are used to explain the disclosure and do not constitute an improper limitation of the disclosure.
[0028] Figure 1 It is the overall structure diagram of the federated learning-based gastric adenocarcinoma pathological image classification system according to an embodiment of the present disclosure;
[0029] Figure 2 It is the detailed structure diagram of the federated learning-based gastric adenocarcinoma pathological image classification according to an embodiment of the present disclosure;
[0030] Figure 3 It is the schematic diagram of multi-scale feature fusion of the federated learning-based gastric adenocarcinoma pathological image classification system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0031] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.
[0032] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further descriptions of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.
[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0034] Embodiment 1
[0035] In an embodiment of the present disclosure, a gastric adenocarcinoma pathological image classification system based on federated learning is provided, including a central server, a plurality of local servers connected to the central server, and a trusted third party; the local server is used to obtain gastric adenocarcinoma pathological images and classify the pathological images based on a classification model; the central server and the plurality of local servers update the classification model through federated learning; the trusted third party is used to generate and split keys;
[0036] Among them, at the beginning stage of federated learning, the central server distributes the global model information to the local servers. The local servers use the keys of a trusted third party to decode the model information, and train to obtain local models on the local servers. After iterating a sufficient number of times, the local models are sent to the central server. After the central server receives the local models sent by all local servers, it restores the keys, decrypts the local models of the local servers and aggregates them to obtain a new global model, which is then sent to each local server as the classification model of the local server.
[0037] As an embodiment, the process of training the classification model is as follows:
[0038] After the local server obtains the local gastric adenocarcinoma pathological images, it classifies the pathological images based on the local model. At the same time, the classification results are submitted to a doctor for review, and then the review results of the doctor are saved to form a local dataset.
[0039] Before federated learning is carried out, the local server uses the local dataset and the reviewed results to train the local model. At the beginning stage of federated learning, through the global model information allocated by the central server, the model is decoded using the keys of a trusted third party, and the local dataset is trained on the local server to obtain a local model. After iterating a sufficient number of times, the local model is sent to the central server.
[0040] After the central server receives and collects the local models sent by all local servers, it restores the keys, decrypts the training results of each local server and aggregates them to obtain a new global model, and before the start of the next round of global training, it sends the new global model to each local server. The trusted third party is used to generate and split the keys.
[0041] The global model has a higher model complexity and a larger scale, while the local model is more suitable for the actual situation of local hospitals and can better adapt to differences in regions, environments, etc.
[0042] As an embodiment, the specific steps of the federated learning include:
[0043] The central server initializes each parameter, and each parameter includes the total number of participating nodes, the number of nodes participating in each round, the global iteration times, the local iteration times, the initialized model, and the training function. The trusted third party generates the main key and encrypts the ciphertext using the main key, and sends it to each local server.
[0044] Specifically, the central server initializes each parameter, including the total number of participating nodes N, the number of nodes participating in each round M, the global iteration times T, the local iteration times P, the initialized model w, and the training function f.
[0045] A trusted third party generates a master key and uses the master key to encrypt the ciphertext, which is then sent to each local server.
[0046] The local server receives the global model sent by the central server and saves it as the local model. The local data is used to train the local model using the master key and ciphertext received from the trusted third party. After the local model training is completed, the local server encrypts the model parameters using the private key generated by the master key and sends the encrypted result and the private key back to the central server.
[0047] The central server selects M out of N nodes as the participating nodes in this round, receives the encrypted results and private keys sent by them, and decrypts them.
[0048] Based on all the results of local training on the local datasets received, the central server aggregates to obtain the global model parameters and sends the aggregated global model to all local servers, thus ending one round of global training.
[0049] Among them, the specific method for optimizing the global model using the local dataset is as follows:
[0050] Based on the weights of the global model, multi-scale image input is used for the training images, and then multi-scale feature fusion is performed on the multi-scale input images. By training the local data, the global model is optimized using stochastic gradient descent and minimizing the loss function.
[0051] Furthermore, the loss function is the cross-entropy loss function:
[0052]
[0053] where n is the number of samples, c is the number of classes, y i,j is the true label of the i-th sample in the j-th class, and p i,j is the predicted probability of the i-th class sample in the j-th class.
[0054] The multi-scale image input uses a Gaussian pyramid and obtains the upper-layer image through convolution and downsampling operations.
[0055] Furthermore, the Gaussian pyramid structure is as follows:
[0056] Based on a Gaussian kernel of size 5×5, the image is convolved, and then the convolved image is downsampled to obtain the upper-layer image G1. This process is repeated on G1 to obtain an even upper-layer image G2, forming a pyramidal image data structure, and a total of three layers of multi-scale images are obtained.
[0057] The multi-scale feature fusion is achieved through a parallel multi-branch network using convolution kernels of different sizes.
[0058] Further, the parallel multi-branch network is as follows:
[0059] Four parallel branch structures are used, namely a 1×1 convolutional layer, a 3×3 convolutional layer, a 5×5 convolutional layer, and a 3×3 max pooling layer, and finally their features are fused.
[0060] Further, the aggregation formula of the global model is:
[0061]
[0062] where M is the number of nodes participating in weight update in this round, is the local model parameter sent by local device i during the (t + 1)-th global model update, and w t+1 is the updated global model.
[0063] As an embodiment, the gastric adenocarcinoma pathological image classification system based on federated learning includes: a number of digital pathology whole slide scanners, a number of workstations, a number of local devices, and a central server.
[0064] The digital pathology whole slide scanner is used to collect gastric adenocarcinoma pathological images, and has functions of scanning, digitizing, compressing, storing, retrieving, and viewing digitized images of human tissue sections. It is an imaging device for observing in vitro human tissue samples.
[0065] A number of digital pathology whole slide scanners in each local medical institution are connected to the same workstation to store the gastric adenocarcinoma pathological images collected by different local digital pathology whole slide scanners. Each local server is connected to a number of workstations, and the workstations are used to upload the gastric adenocarcinoma pathological images collected by different ultrasound devices to the local server.
[0066] The workstation includes a data acquisition module and a pre-classification module;
[0067] The data collection module of the workstation is to improve the portability of reading images stored in different scanners and increase the image storage space. Different image acquisition cards are used to collect the gastric adenocarcinoma pathological images in different scanners into the workstation, so that the pathological images can be more conveniently retained and read. The workstation is connected to different scanners, and the scanners are used to scan and collect the gastric adenocarcinoma pathological images in different ultrasound devices into the workstation.
[0068] The data collection module of the workstation can also be used to read the pathological images entered by the administrator.
[0069] The pre-classification module of the workstation can call the local server to pre-classify the pathological images collected by the scanner and the pathological images entered by the administrator. Then, the pathologist reviews the pre-classification results, adds category labels to them, forms a local dataset, and uploads it to the local server.
[0070] Specifically, based on the pathological images, the pathological images stored in the workstation are divided into nine categories: stage 0, stage IA, stage IB, stage IIA, stage IIB, stage IIIA, stage IIIB, stage IIIC, and stage IV.
[0071] As an example, a certain hospital arranges a local server to connect all scanners and workstations within the hospital.
[0072] The local server includes a model training module and a result output module. The model training module of the local server is used to read the global model sent by the central server, obtain the training set stored in the workstation, combine them into a local training set, use the local training set to train the classification model, form a local model, and send the model back to the central server.
[0073] Taking the above hospital as an example, the pathological images of the local training set are used as the bottom-layer image G0, which is then sent into the Gaussian pyramid. A Gaussian kernel of size 5×5 is used to perform convolution on it, and then the convolved image is downsampled to obtain the upper-layer image G1. This process is repeated for G1 to obtain an even more upper-layer image G2, forming a pyramidal image data structure, and a total of three layers of multi-scale images are obtained.
[0074] Taking the bottom-layer image G0 as an example, four parallel branch structures are used, namely 1×1 convolution, 3×3 convolution, 5×5 convolution, and 3×3 max pooling. Finally, multi-scale feature fusion is performed on the four channels. The same operations are performed on the G1 and G2 layer images. Finally, the features of the three layers of images are fused and sent to the local server for training. After the training is completed, the updated model of the local server is sent back to the central server. The remaining hospital centers follow the same classification model training steps. Finally, the central server distributes the updated classification model to each hospital center again.
[0075] Specifically:
[0076] The central server maintains a global central model f with initial weights w t , and the central server encrypts the initial weights and sends them to each local server.
[0077] Each local device i receives the initial weights w t, for the local dataset, first perform multi-scale input based on the Gaussian pyramid, and perform multi-scale feature fusion on each layer of the image. Then, based on local objective minimization, use the mini-batch stochastic gradient descent algorithm with the smallest local learning rate to perform the training step on the local private data, and optimize the model by minimizing the cross-entropy loss of classification;
[0078] If the local training is completed, each hospital center sends their updated model updates back to the central server.
[0079] Finally, the central server receives the updated models from all local servers and decrypts them. Select M local models, aggregate them, and calculate a new global model w t+1 , and update each parameter of the global model f;
[0080]
[0081] Taking the bottom-layer Gaussian image feature g0 as an example, construct the multi-scale feature g0 through four operations: 1×1 convolution, 3×3 convolution, 5×5 convolution, and 3×3 max pooling. Let the 1×1 convolution be layer c0, the 3×3 convolution be layer c1, the 5×5 convolution be layer c2, and the 3×3 max pooling be layer c3, then:
[0082]
[0083] Among them, let the bottom-layer Gaussian image feature be g0, the upper-layer Gaussian image feature be g1, and the upper-upper-layer Gaussian image feature be g2;
[0084] The feature fusion step based on the multi-scale image is as follows:
[0085] g = (g0 + g1 + g2) / 3
[0086] Among them, are the features obtained by the bottom-layer Gaussian image through layers c0, c1, c2, and c3 respectively.
[0087] So far, the data feature fusion step of the local server training of a hospital is completed.
[0088] The update and iteration process of federated learning is formed between the local server and the central server. This process is continuously repeated. After T iterations, the learning process ends, and the optimal model is reached on the central server and then sent to the local servers of each hospital.
[0089] This embodiment trains a gastric adenocarcinoma pathological image classification system with strong robustness and high accuracy by fusing the common features of pathological images based on multiple local devices and different scanner sources.
[0090] Embodiment 2
[0091] In one embodiment of the present disclosure, an electronic device is provided, including: a memory for non-temporarily storing computer-readable instructions; and a processor for executing the computer-readable instructions, wherein when the computer-readable instructions are run by the processor, the following steps are performed: obtaining a gastric adenocarcinoma pathological image; classifying the gastric adenocarcinoma pathological image by using a classification model updated through federated learning mentioned in Embodiment 1.
[0092] In this embodiment, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0093] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.
[0094] In the implementation process, each step of the above method may be completed by the integrated logic circuit in the hardware of the processor or the instructions in the form of software.
[0095] Embodiment 3
[0096] In one embodiment of the present disclosure, a storage medium is provided, including non-temporarily storing computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the following steps are performed: obtaining a gastric adenocarcinoma pathological image; classifying the gastric adenocarcinoma pathological image by using a classification model updated through federated learning mentioned in Embodiment 1.
[0097] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.
[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps for implementing the functions specified in one block or a plurality of blocks.
[0099] Although the specific embodiments of the present disclosure have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that, based on the technical solutions of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative efforts are still within the scope of protection of the present disclosure.
Claims
1. A gastric adenocarcinoma pathology image classification system based on federated learning, characterized in that: It includes a central server, multiple local servers connected to the central server, and a trusted third party; The local server is used to obtain gastric adenocarcinoma pathological images and classify the pathological images based on the classification model; the central server and multiple local servers update the classification model through federated learning; the trusted third party is used to generate and split the key; Among them, at the beginning stage of federated learning, the central server distributes global model information to local servers. The local servers use the keys of trusted third parties to decode the model information and train local models on the local servers. After a sufficient number of iterations, the local models are sent to the central server. After receiving the local models sent by all local servers, the central server restores the key, decrypts the local models of the local servers and aggregates them to obtain a new global model and send it to each local server as the classification model of the local server.
2. The gastric adenocarcinoma pathology image classification system based on federated learning according to claim 1, characterized in that: The specific steps of the federated learning include: the central server initializes various parameters, including the total number of participating nodes, the number of nodes participating in each round, the number of global iterations, the number of local iterations, the initialization model and the training function; The trusted third party generates a master key, encrypts the ciphertext using the master key, and sends the ciphertext to each local server.
3. The gastric adenocarcinoma pathology image classification system based on federated learning as claimed in claim 2, characterized in that: The local server receives the global model information sent by the central server and saves it as a local model, and trains the local model using the master key and ciphertext received from a trusted third party; after the local model training is completed, the local server uses the private key generated by the master key to encrypt the model parameters, and returns the encryption result and the private key to the central server; the central server selects multiple nodes from a number of nodes as participating nodes in this round, receives the encryption result and private key sent, and decrypts them; the central server aggregates the global model parameters based on all the results of local training based on the local data set received, and sends the aggregated global model to all local servers, and this round of global training ends.
4. The gastric adenocarcinoma pathology image classification system based on federated learning as claimed in claim 3, characterized in that: The specific method of using local data sets to optimize the global model is as follows: based on the weights of the global model, use multi-scale image input for training images, and then perform multi-scale feature fusion on the multi-scale input images. By training local data, the global model is optimized using stochastic gradient descent and minimizing the loss function.
5. The gastric adenocarcinoma pathology image classification system based on federated learning as claimed in claim 4, characterized in that: The multi-scale image input uses a Gaussian pyramid, and uses convolution and downsampling operations to obtain a higher layer of image. The multi-scale feature fusion is achieved through a parallel multi-branch network using convolution kernels of different sizes.
6. The gastric adenocarcinoma pathology image classification system based on federated learning according to claim 1, characterized in that: The aggregation formula for obtaining the global model is: Among them, M is the number of nodes participating in the weight update in this round, is the local model parameter sent by local server i at the t+1th global model update, ω t+1 is the updated global model.
7. The gastric adenocarcinoma pathology image classification system based on federated learning according to claim 1, characterized in that: Each local server is connected to a number of workstations, and the workstations are used to upload gastric adenocarcinoma pathological images collected by different ultrasound devices to the local server.
8. The gastric adenocarcinoma pathology image classification system based on federated learning according to claim 7, characterized in that: The workstation is connected to different scanners, and the scanners are used to scan and collect gastric adenocarcinoma pathological images from different ultrasound devices into the workstation.
9. An electronic device, characterized in that: include: A memory for non-temporarily storing computer-readable instructions; and a processor for executing the computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the following steps are performed: acquiring a gastric adenocarcinoma pathology image; and classifying the gastric adenocarcinoma pathology image using a classification model updated by federated learning.
10. A storage medium, characterized in that: The method comprises non-temporarily storing computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the following steps are performed: obtaining a gastric adenocarcinoma pathology image; and classifying the gastric adenocarcinoma pathology image using a classification model updated by federated learning.