Pulmonary nodule detection system and method based on federated learning and target detection algorithm

Abstract

The invention discloses a pulmonary nodule detection system and method based on federated learning and a target detection algorithm, and relates to the technical field of medical image detection and auxiliary diagnosis, the system comprises a plurality of medical institution clients and a cloud central server, the central server initializes to generate a global model, and deploys the global model to each client; each client uses a local CT data set to train a global model, model parameters are updated through training until loss is minimized so as to complete training, a local model is generated, and the updated model parameters are uploaded to the central server; wherein the total loss of the local model is the dynamic weighted sum of various losses; the central server adopts a dynamic weighted average method to aggregate the updated model parameters of all the clients, generates an updated global model, then deploys the updated global model, and gradually optimizes the global model through multiple times of global iteration updating; and each client identifies the to-be-detected CT image by using the locally deployed optimized global model, and outputs a more accurate pulmonary nodule identification result.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image detection and auxiliary diagnosis, and particularly to a pulmonary nodule detection system and method based on federated learning and object detection algorithms. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Pulmonary nodules are a complex class of diseases, including various different types of abnormal growths, and their diagnosis and treatment are crucial for the prognosis of patients. Early and accurate detection of pulmonary nodules is of great significance for selecting appropriate treatment plans and improving the prognosis of patients. Among them, CT imaging, as a non-invasive imaging examination method, can provide detailed images of pulmonary soft tissues and is currently the main tool for detecting and evaluating pulmonary nodules. However, the interpretation of CT images is extremely complex and requires rich professional knowledge and experience. Misdiagnosis may bring serious consequences. Therefore, improving the accuracy and efficiency of pulmonary nodule detection and recognition is not only an engineering challenge but also an urgent medical need.

[0004] Traditional pulmonary nodule detection and recognition mostly rely on the manual judgment of professionals with rich experience, which is highly subjective and has poor recognition accuracy. With the development of machine learning and deep learning technologies, significant progress has been made in the field of pulmonary nodule object detection, but there are still the following key problems:

[0005] (1) Data privacy issues: Traditional centralized learning methods require a large amount of sensitive medical data to be transmitted to a central server for model training, which brings potential data leakage risks and violates strict privacy regulations.

[0006] (2) Poor model training effect: There are large differences in data distributions among different medical institutions. In scenarios facing large changes in data distribution and complex tasks, it is difficult to balance the training performance and training stability of the model, and the finally trained model has poor performance.

[0007] (3) Data heterogeneity: The datasets of a single institution usually have limitations, with limited data volume and possible biases, resulting in insufficient generalization ability of the model. Moreover, there are large differences in data distributions among different medical institutions, which further affects the universality of the model.

[0008] (4) High computational resource requirements: Deep learning models usually require a large amount of computational resources and storage space, which is a major challenge for small medical institutions with limited resources.

[0009] (5) Slow CT image processing speed: CT has high resolution, but the traditional detection model has a slow detection speed, and model training greatly affects the timeliness of detection. Summary of the Invention

[0010] To solve the deficiencies of the above-mentioned existing technologies, the present invention provides a pulmonary nodule detection system and method based on federated learning and object detection algorithms. By adopting the method of federated learning and combining with the object detection algorithm - YOLOv12 model, the privacy of patients is protected by sharing model updates instead of raw data among multiple medical institutions, the universality of the model is improved, and at the same time, the loss function during local training of the shared model is improved to further optimize the local model, effectively improving the accuracy and efficiency of pulmonary nodule detection in CT images.

[0011] In the first aspect, the present invention provides a pulmonary nodule detection system based on federated learning and object detection algorithms.

[0012] A pulmonary nodule detection system based on federated learning and object detection algorithms includes multiple medical institution clients and a central server in the cloud;

[0013] Deployed in the central server: A global model initialization module for initializing and generating a global pulmonary nodule object detection model and deploying it to each client;

[0014] A global model update module for aggregating the model parameters updated by all clients using the dynamic weighted average method to generate an updated global model, and then deploying the updated model to each client. After multiple global iterative updates, the global model is gradually optimized;

[0015] Deployed in each medical institution client: A local model optimization module for obtaining local CT images to construct a CT dataset, training the global model deployed on its own using the local CT dataset, updating the model parameters through multiple iterative trainings until the loss function is minimized to complete the training, generating a local pulmonary nodule object detection model, and uploading the updated model parameters after training to the central server; among them, the total loss of the local model is the dynamic weighted sum of multiple losses;

[0016] A pulmonary nodule detection module for obtaining local CT images to be detected, preprocessing them, and inputting them into the optimized global model deployed locally to output the pulmonary nodules in the recognized images.

[0017] A further technical solution is that the total loss of the local pulmonary nodule object detection model is the dynamic weighted sum of coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss;

[0018] Among them, each time the local model is iteratively trained, according to the global model parameters deployed in this iteration, calculate the gradient norms of the losses of the local model and the target gradient norm, and dynamically adjust the weights of the losses according to the difference between the gradient norm of the current loss and the target gradient norm, so as to adjust the final total loss.

[0019] In a second aspect, the present invention provides a lung nodule detection method based on federated learning and object detection algorithms.

[0020] A lung nodule detection method based on federated learning and object detection algorithms includes:

[0021] The central server initializes and generates a global lung nodule object detection model and deploys it to the client of each medical institution;

[0022] For each client, obtain local CT images to construct a CT dataset, use the local CT dataset to train the globally deployed model, and after multiple iterative trainings, update the model parameters until the loss function is minimized, generate a local lung nodule object detection model, and upload the updated model parameters after training to the central server; among them, the total loss of the local model is the dynamic weighted sum of multiple losses;

[0023] The central server uses the dynamic weighted average method to aggregate the updated model parameters of all clients, generates an updated global model, and then deploys the updated model to each client. After multiple global iterative updates, the global model is gradually optimized;

[0024] For each client, obtain local CT images to be detected, and after preprocessing, input them into the local model deployed locally to output the lung nodules in the recognized images.

[0025] In a third aspect, the present invention further provides an electronic device, including: a memory for storing executable instructions; a processor, when executing the executable instructions stored in the memory, implementing the above-mentioned lung nodule detection system based on federated learning and object detection algorithms, or executing the steps of the above-mentioned lung nodule detection method based on federated learning and object detection algorithms.

[0026] In a fourth aspect, the present invention further provides a computer-readable storage medium storing executable instructions, which are used to cause a processor to implement the above-mentioned lung nodule detection system based on federated learning and object detection algorithms, or execute the steps of the above-mentioned lung nodule detection method based on federated learning and object detection algorithms when executing the executable instructions.

[0027] In a fifth aspect, the present invention also provides a computer program product, which includes executable instructions stored in a computer-readable storage medium; wherein, when a processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements the above-mentioned lung nodule detection system based on federated learning and object detection algorithms, or executes the steps of the above-mentioned lung nodule detection method based on federated learning and object detection algorithms.

[0028] The above one or more technical solutions have the following beneficial effects:

[0029] 1. The present invention provides a lung nodule detection system and method based on federated learning and object detection algorithms, which is implemented by adopting the federated learning method and combining with the existing latest object detection algorithm - YOLOv12 model. Among them: through the federated learning architecture, model updates rather than raw data are shared among multiple medical institutions, thus effectively protecting patient privacy and avoiding the risk of sensitive data leakage; this method integrates diverse data from different medical facilities, significantly improving the generalization ability and robustness of the model, and solving the bias problem that may exist in the dataset of a single institution; the distributed training mechanism reduces the computational burden on the central server, enabling small medical institutions with limited resources to efficiently participate in model training and reducing the demand for high computing resources; at the same time, using the powerful real-time object detection ability of YOLOv12 and combining with the loss function (including coordinate loss, size loss, confidence loss, and classification loss) during local training of the improved shared model to further optimize the training of the local model, ensuring high performance and high-precision lung nodule detection and classification, and effectively improving the accuracy and efficiency of lung nodule detection in CT images.

[0030] 2. In the lung nodule detection system and method based on federated learning and object detection algorithms proposed by the present invention, during the training process of the local model, multiple losses are introduced: coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss. The dynamic weighted sum of multiple losses is used as the total loss for iterative training. By setting multiple loss functions, the comprehensiveness of the final detected lung nodules can be ensured, realizing accurate detection of the lung nodules themselves and their positions, sizes, and types, and improving the detection accuracy; through dynamic weight optimization, the local model can dynamically adjust the loss weights according to the training conditions corresponding to different losses, avoiding overfitting or underfitting of certain tasks, and being able to achieve a better balance between model performance and training stability, avoiding the limitations of traditional static weight allocation.

[0031] The advantages of the additional aspects of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings of the specification, which form a part of the present invention, are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0033] Figure 1 It is the overall architecture diagram of the pulmonary nodule detection system based on federated learning and object detection algorithm described in the embodiments of the present invention;

[0034] Figure 2 It is the schematic diagram of the result of pulmonary nodule detection using the optimized global model deployed locally in the embodiments of the present invention. Detailed implementation manners

[0035] It should be noted that the following detailed description is exemplary and is only for describing the specific implementation manners, aiming to provide further explanation of the present invention and not intended to limit the exemplary embodiments according to the present invention. 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 invention belongs. In addition, it should also 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.

[0036] Embodiment 1

[0037] This embodiment provides a pulmonary nodule detection system based on federated learning and object detection algorithm. As Figure 1 shown, the system includes multiple medical institution clients and a central server in the cloud. Each client maintains its local CT dataset, uses the YOLOv12 model to train on its local dataset, and generates model update parameters. The central server is responsible for receiving and aggregating the model update parameters uploaded by each client, and uses the dynamic weighted average method to integrate all update parameters to generate and optimize the global model, and then redistributes it to each client. In each client, the obtained CT images to be detected are preprocessed by standardization to improve the robustness of the model, and then the preprocessed CT images are input into the optimized global model deployed locally, and the pulmonary nodules in the recognized images are output.

[0038] In the pulmonary nodule detection system based on federated learning and object detection algorithm proposed in this embodiment, a global model initialization module and a global model update module are deployed in the central server, and a local model optimization module and a pulmonary nodule detection module are deployed in the medical institution clients, where:

[0039] (1) The global model initialization module is used to initialize and generate a global pulmonary nodule object detection model and deploy it to each client.

[0040] In this embodiment, the central server initializes a global YOLOv12 model with parameters After that, the central server distributes the model parameters of the current global model to each client n ∈ N, where k represents the number of update iterations.

[0041] (2) The local model optimization module is used to obtain local CT images to construct a CT dataset, train the globally deployed model with the local CT dataset, update the model parameters through multiple iterations of training until the loss function is minimized to complete the training, generate a local lung nodule target detection model, and upload the updated model parameters after training to the central server.

[0042] In this embodiment, each client maintains its own local CT dataset, and after receiving the global model, loads the local dataset and initializes the local model After that, the client performs multiple rounds of training (such as I rounds) on the local dataset, and updates the model parameters using the stochastic gradient descent method in each round The update formula is where η is the learning rate and L is the loss function; after the training is completed, the client sends the updated model parameters θ k,new back to the central server.

[0043] In the above process, the globally deployed model is continuously iteratively trained and updated using the local CT dataset. Among them, to comprehensively optimize the lung nodule detection model, this embodiment introduces multiple loss calculation methods, that is, the total loss during the iterative training process is the dynamic weighted sum of the coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss. Among them, the coordinate loss reflects the difference between the center point coordinates of the predicted bounding box and the center point coordinates of the true bounding box. By minimizing this difference, the model can accurately locate the lung nodule; the size loss reflects the difference between the width and height of the predicted bounding box and the width and height of the true bounding box. By minimizing this difference, the model can more accurately estimate the size of the lung nodule; the confidence loss reflects the difference between the confidence score of the predicted bounding box and the true confidence score. By minimizing this difference, the model can more accurately determine whether there is a lung nodule; the non-object confidence loss reflects the difference between the confidence score of the predicted bounding box in the background area and the true confidence score. By minimizing this difference, the model can reduce false alarms in the background area; the classification loss reflects the difference between the predicted class probability distribution and the true class probability distribution. By minimizing this difference, the model can more accurately identify the type of lung nodule. By setting the above multiple loss functions, the comprehensiveness of the final detection of lung nodules can be ensured, and the accurate detection of the lung nodule itself, as well as the position, size, and type of the lung nodule, can be achieved.

[0044] (1) The coordinate loss is as follows: The CT image is divided into several grid cells, the predicted bounding boxes in each grid cell are aggregated, and based on the target objects detected in the predicted bounding boxes within all grid cells, as well as the center coordinates of the ground truth bounding box and the predicted bounding box, the coordinate loss is formed; the calculation formula for this coordinate loss is:

[0045]

[0046] Among them, λ coord is the weight factor, used to increase the importance of coordinates in the loss; S 2 represents the grid size, that is, the image is divided into a grid of S×S; i represents the index of the current grid cell; B represents the number of predicted bounding boxes in each grid cell; j represents the index of the current bounding box; is the indicator function, which is 1 if the bounding box j in the grid cell i contains the target object, otherwise 0; x i ,y i are the center point coordinates of the predicted bounding box, x′ i ,y′ i are the center point coordinates of the ground truth bounding box.

[0047] (2) The size loss is as follows: The CT image is divided into several grid cells, the predicted bounding boxes in each grid cell are aggregated, and based on the target objects detected in the predicted bounding boxes within all grid cells, as well as the width and height of the ground truth bounding box and the predicted bounding box, the size loss is formed; the calculation formula for this size loss is:

[0048]

[0049] Among them, λ size is the weight factor, used to increase the importance of size in the loss; is the indicator function, which is 1 if the bounding box j in the grid cell i contains the target object, otherwise 0; w i ,h i represent the width and height of the predicted bounding box, w′ i ,h′ i represent the width and height of the ground truth bounding box, and the square root is used to reduce the influence of large size errors.

[0050] (3) The confidence loss is as follows: The CT image is divided into several grid cells, the predicted bounding boxes in each grid cell are aggregated, and based on the presence of the target objects detected in the predicted bounding boxes within all grid cells, as well as the confidence scores of the ground truth bounding box and the predicted bounding box, the confidence loss is formed; the calculation formula for this confidence loss is:

[0051]

[0052] Among them, λ conf is the weight factor, used to adjust the weight of the confidence loss; S 2 represents the grid size, that is, the image is divided into a grid of S×S; i represents the index of the current grid cell; B represents the number of bounding boxes predicted for each grid cell; j represents the index of the current bounding box; is the indicator function, which is 1 if the bounding box j in the grid cell i contains the target object, otherwise 0; C i is the confidence score of the predicted bounding box, estimating the intersection over union between the predicted bounding box and the ground truth bounding box; C′ i is the confidence score of the ground truth bounding box.

[0053] (4) The non-object confidence loss is as follows: The CT image is divided into several grid cells, and the predicted bounding boxes in each grid cell are aggregated. According to the non-existence of the detected target object in the predicted bounding boxes in all grid cells, as well as the confidence scores of the ground truth bounding box and the predicted bounding box, the non-object confidence loss is constructed; this non-object confidence loss is:

[0054]

[0055] Among them, λ noconf is the weight factor, used to adjust the weight of the confidence loss; is the indicator function, which is 1 if the bounding box j in the grid cell i does not contain the target object, otherwise 0; C i is the confidence score of the predicted bounding box, estimating the intersection over union between the predicted bounding box and the ground truth bounding box; C′ i is the confidence score of the ground truth bounding box.

[0056] (5) The classification loss is as follows: The CT image is divided into several grid cells, and according to the detected target object in the grid cell and the predicted probability and the true probability of the detected target object category, the classification loss is constructed; the calculation formula of this classification loss is:

[0057]

[0058] Among them, λ class represents the weight size, used to adjust the weight of the classification loss; S 2 represents the grid size, that is, the image is divided into a grid of S×S; i represents the index of the current grid cell; is the indicator function, which is 1 if the grid cell i does not contain the target object, otherwise 0; p i (c) represents the predicted probability of the grid cell i for the category c; p′ i (c) represents the true probability of the grid cell i for the category c.

[0059] Furthermore, to avoid the limitations of static weight allocation during each local model training process, this embodiment adopts a dynamic weight factor generation method to aggregate different types of losses, that is: initialize the initial weight λ of each loss of the local model t = 1, and this initial weight λ t is the weight of the t-th unweighted loss L t , where t ∈ T, and T = {coord, size, conf, noconf, class}; during the iterative update process of the local model, according to the global model parameters θ g deployed by the central server to the local client, calculate the gradient norm of each loss of the local model Furthermore, the target gradient norm can be calculated based on the gradient norms of all losses This target gradient norm is the geometric mean of the gradient norms of all losses. According to the difference between the gradient norm of the current loss and the target gradient norm, dynamically adjust the weight λ of each loss t , and the update formula for this loss weight is: where p = 0.8 is a hyperparameter used to control the intensity of adjustment. Finally, calculate the total loss function of this iterative training using the adjusted weights to comprehensively optimize the pulmonary nodule detection model.

[0060] Through the above dynamic weight optimization, the local model can dynamically adjust the loss weights according to the training situations corresponding to different losses, avoid overfitting or underfitting of certain tasks, and can achieve a better balance between model performance and training stability, avoiding the limitations of traditional static weight allocation.

[0061] (3) Global model update module, which is used to aggregate the model parameters updated by all clients using the dynamic weighted average method to generate an updated global model, and then deploy the updated model to each client. After multiple global iterative updates, gradually optimize the global model.

[0062] In this embodiment, a small part of the data is reserved in the central server as a validation dataset to validate the updated models uploaded by all clients, and the detection accuracy rate of each client model obtained is used as the validation score s n , and then, according to this validation score, aggregate the models updated by all clients using the dynamic weighted average method to update and optimize the global model and generate updated global model parameters The formula is: where N is the number of all clients.

[0063] Furthermore, continuously iterate the process of the above steps S1 to S3, and gradually optimize the global model through multiple global rounds.

[0064] (4) The lung nodule detection module is used to obtain the local CT image to be detected, and after preprocessing, input it into the optimized global model deployed locally, and output the lung nodules in the recognized image.

[0065] In this embodiment, the local CT image to be detected is obtained, and preprocessing such as adjusting the resolution and normalizing the pixel values of the CT image is performed. After preprocessing, the image is input into the optimized global model deployed locally, and the lung nodules in the recognized image are output. The final recognition result is as Figure 2 shown.

[0066] As an implementation method, the model deployed locally is regularly verified and evaluated. The model is used to process the new data added each month in the actual scenario. When the detection accuracy rate is lower than the set value, such as 90%, according to the verification and evaluation results, the model is trained and updated in a timely manner to ensure the continuous improvement of its performance.

[0067] Embodiment 2

[0068] This embodiment provides a lung nodule detection method based on federated learning and object detection algorithm, which specifically includes the following steps:

[0069] Step S1: The central server initializes and generates a global lung nodule object detection model, and deploys it to each medical institution client;

[0070] Step S2: For each client, obtain the local CT image to construct a CT dataset, and use the local CT dataset to train the globally deployed model of each. After multiple iterative trainings, update the model parameters until the loss function is minimized, generate a local lung nodule object detection model, and upload the updated model parameters after training to the central server; among them, the total loss of the local model is the dynamic weighted sum of multiple losses;

[0071] Step S3: The central server uses the dynamic weighted average method to aggregate the updated model parameters of all clients, generate an updated global model, and then deploy the updated model to each client. After multiple global iterative updates, gradually optimize the global model;

[0072] Step S4: For each client, obtain the local CT image to be detected, and after preprocessing, input it into the local model deployed locally, and output the lung nodules in the recognized image.

[0073] Furthermore, the total loss of the local lung nodule object detection model is the dynamic weighted sum of the coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss;

[0074] Among them, each time the local model is iteratively trained, according to the global model parameters deployed in this iteration, calculate the gradient norms of the losses of the local model and the target gradient norm, and dynamically adjust the weights of the losses according to the difference between the gradient norm of the current loss and the target gradient norm, so as to adjust the final total loss.

[0075] Embodiment III

[0076] This embodiment provides an electronic device, including: a memory for storing executable instructions; a processor for implementing the above method provided in this embodiment when executing the executable instructions stored in the memory.

[0077] Embodiment IV

[0078] This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, will cause the processor to execute the above method provided in this embodiment.

[0079] Embodiment V

[0080] This embodiment provides a computer program product, which includes executable instructions, and the executable instructions are computer instructions; the executable instructions are stored in a computer-readable storage medium. When the processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the electronic device is caused to execute the above method provided in this embodiment.

[0081] The steps involved in Embodiments II to V above correspond to those in Embodiment I, and the specific implementation manners can be referred to the relevant description part of Embodiment I. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to execute any method in the present invention.

[0082] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to implement. The present invention is not limited to any specific combination of hardware and software.

[0083] The above are only the preferred embodiments of the present invention. Although the specific embodiments of the present invention are described in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications or deformations that can be made without creative efforts based on the technical solution of the present invention are still within the protection scope of the present invention.

Claims

1. A lung nodule detection system based on federated learning and object detection algorithms, characterized in that, It includes multiple medical institution clients and a central server in the cloud; In the central server, there is a global model initialization module, which is used to initialize and generate a global pulmonary nodule target detection model and deploy it to each client; A global model update module, which is used to aggregate the model parameters updated by all clients by using the dynamic weighted average method to generate an updated global model, and then deploy the updated model to each client. After multiple global iterative updates, the global model is gradually optimized; In each medical institution client, there is a local model optimization module, which is used to obtain local CT images to construct a CT dataset, and use the local CT dataset to train the global model deployed by each of them. After multiple iterative trainings, the model parameters are updated until the loss function is minimized to complete the training, generating a local pulmonary nodule target detection model, and uploading the updated model parameters after training to the central server; among them, the total loss of the local model is the dynamic weighted sum of multiple losses; A pulmonary nodule detection module, which is used to obtain local CT images to be detected, and after preprocessing, input them into the optimized global model deployed locally to output the pulmonary nodules in the recognized images.

2. The pulmonary nodule detection system based on federated learning and object detection algorithm according to claim 1, wherein, The total loss of the local pulmonary nodule target detection model is the dynamic weighted sum of coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss; Among them, each time the local model is iteratively trained, according to the global model parameters deployed in this iteration, calculate the gradient norm of each loss of the local model and the target gradient norm, and dynamically adjust the weights of each loss according to the difference between the current loss gradient norm and the target gradient norm to adjust the final total loss.

3. The pulmonary nodule detection system based on federated learning and object detection algorithm according to claim 2, characterized in that, The coordinate loss is as follows: The CT image is divided into several grid cells, and according to the target objects detected in the predicted bounding boxes within all grid cells, as well as the center coordinates of the true bounding box and the predicted bounding box, the coordinate loss is formed; The size loss is as follows: The CT image is divided into several grid cells, and according to the target objects detected in the predicted bounding boxes within all grid cells, as well as the width and height of the true bounding box and the predicted bounding box, the size loss is formed.

4. The pulmonary nodule detection system based on federated learning and object detection algorithm according to claim 2, wherein The confidence loss is as follows: The CT image is divided into several grid cells, and according to the presence of the target objects detected in the predicted bounding boxes within all grid cells, as well as the confidence scores of the true bounding box and the predicted bounding box, the confidence loss is formed; The non-object confidence loss is as follows: The CT image is divided into several grid cells, and according to the absence of the target objects detected in the predicted bounding boxes within all grid cells, as well as the confidence scores of the true bounding box and the predicted bounding box, the non-object confidence loss is formed.

5. The lung nodule detection system based on federated learning and object detection algorithm according to claim 2, wherein, The classification loss is as follows: The CT image is divided into several grid cells, and according to the target objects detected within all grid cells and the predicted probability and true probability of the detected target object categories, the classification loss is formed.

6. A lung nodule detection method based on federated learning and object detection algorithms, characterized in that, It includes: The central server initializes and generates a global pulmonary nodule target detection model and deploys it to each medical institution client; For each client, obtain local CT images to construct a CT dataset, and use the local CT dataset to train the globally deployed model on each client. After multiple iterative trainings, update the model parameters until the loss function is minimized to generate a local lung nodule target detection model, and upload the updated model parameters after training to the central server. Among them, the total loss of the local model is the dynamic weighted sum of multiple losses. The central server aggregates the updated model parameters of all clients using the dynamic weighted average method to generate an updated global model, and then deploys the updated model to each client. After multiple global iterative updates, the global model is gradually optimized. For each client, obtain the local CT image to be detected. After preprocessing, input it into the locally deployed local model to output the lung nodules in the recognized image.

7. The pulmonary nodule detection method based on federated learning and object detection algorithm according to claim 6, wherein The total loss of the local lung nodule target detection model is the dynamic weighted sum of the coordinate loss, size loss, confidence loss, non-object confidence loss, and classification loss. Among them, each time the local model is iteratively trained, according to the global model parameters deployed in this iteration, calculate the gradient norm of each loss of the local model and the target gradient norm. According to the difference between the current loss gradient norm and the target gradient norm, dynamically adjust the weights of each loss to adjust the final total loss.

8. An electronic device, characterized in that, It includes: A memory for storing executable instructions; A processor, when executing the executable instructions stored in the memory, implements a lung nodule detection system based on federated learning and target detection algorithms as described in any one of claims 1-5, or executes the steps of a lung nodule detection method based on federated learning and target detection algorithms as described in claims 6-7.

9. A computer-readable storage medium, characterized in that, Stored with executable instructions, which are used to cause the processor to implement a lung nodule detection system based on federated learning and target detection algorithms as described in any one of claims 1-5, or execute the steps of a lung nodule detection method based on federated learning and target detection algorithms as described in claims 6-7 when executing the executable instructions.

10. A computer program product, characterized in that, The computer program product includes executable instructions, and the executable instructions are stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements a lung nodule detection system based on federated learning and target detection algorithms as described in any one of claims 1-5, or executes the steps of a lung nodule detection method based on federated learning and target detection algorithms as described in claims 6-7.

Images