Lung CT image processing system and method based on end-side cloud architecture

Through the lung CT imaging processing system based on the end-edge cloud architecture, combined with deep learning and edge computing, the problem of inaccuracy and efficiency in ARDS diagnosis and treatment is solved, and the rapid and accurate etiology analysis and treatment plan are achieved, improving the provision of improved etiology Treatment effects of ARDS patients.

CN119048457APending Publication Date: 2024-11-29浪潮智能终端有限公司
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411119354.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The prior art has low accuracy and analysis efficiency in early diagnosis and treatment of ARDS and cannot meet actual needs.

Method used

Using the lung CT image processing system based on the end-edge cloud architecture, combined with edge computing and deep learning technology, the edge server, terminal equipment and cloud computing communicate with each other, and the acceleration card accelerates the operation of the lungs of ARDS patients is achieved quickly segmented and analyzed. CT imaging data provides high-accurate causes and optimal treatment options.

Benefits of technology

It has improved the monitoring and treatment effect of ARDS patients, met the real-time and accurate data processing needs in the medical field, improved the treatment level, and provided technical support for other medical fields.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048457A_ABST
    Figure CN119048457A_ABST
Patent Text Reader

Abstract

The invention discloses a lung CT image processing system and method based on an end-side cloud architecture, and belongs to the technical field of computer science. Comprising an edge server, terminal equipment and cloud computing, and the terminal equipment comprises a model deployment module, a file processing module, an information processing and display module, a server communication module and an image segmentation and analysis module; the edge server comprises an edge acceleration module; the cloud computing comprises a database and a GPU; and the edge server communicates with the terminal equipment and the cloud computing. According to the method, the edge calculation and the deep learning technology are combined, the deep learning model is realized on the edge equipment, and the quick response and high-efficiency capability are realized, so that the requirements of the medical field on real-time and accurate data processing are better met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a lung CT image processing system and method based on an edge-cloud architecture, and belongs to the fields of computer science, medical engineering and deep learning technology. Background Art

[0002] ARDS, or Acute Respiratory Distress Syndrome, is a serious lung disease characterized by acute onset of respiratory distress, hypoxemia, and increased alveolar permeability. ARDS is usually caused by a variety of reasons, including infection (such as pneumonia), trauma, systemic infection, severe trauma or surgery, and some other factors. The main characteristics of ARDS include damage to the alveoli and capillaries, which causes the alveoli to be filled with fluid, making it difficult for oxygen to enter the blood through the alveoli, causing respiratory distress. ARDS patients usually require mechanical ventilation to maintain adequate oxygenation.

[0003] At present, there are deep learning algorithms used in the early diagnosis of ARDS: by performing convolutional neural network (CNN) analysis on clinical images, including X-rays and CT scans, to improve the sensitivity to subtle pathological changes, so as to further improve the accuracy of early diagnosis of ARDS; there are genomics applications in personalized treatment plans for ARDS: through systematic analysis of patient genetic data, we try to determine the treatment response associated with a specific genotype, thereby promoting the practice of personalized medicine in ARDS management; there are machine learning algorithms to establish a predictive model for the course of ARDS patients and treatment response: by integrating clinical data, early intervention decision support is provided to improve patient prognosis. There is also a respiratory support technology based on edge computing and intelligent control, which aims to improve the ventilation effect of ARDS patients. By monitoring the patient's breathing pattern in real time, the system can adjust the ventilation parameters to meet the needs of personalized treatment to the greatest extent. The current solution has problems with low accuracy and analysis efficiency and cannot meet current actual needs. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a lung CT image processing system and method based on an edge-cloud architecture, which can quickly segment and analyze the lung CT image data of ARDS patients to provide highly accurate causes and optimal treatment plans.

[0005] It includes an edge server, a terminal device, and cloud computing. The terminal device includes: a model deployment module, a file processing module, an information processing and display module, a server communication module, and an image segmentation and analysis module; the edge server includes an edge acceleration module; the cloud computing includes a database and a GPU; the edge server communicates with the terminal device and the cloud computing; The edge acceleration module realizes the rapid operation of the model through the acceleration card; the model deployment module is configured to install the medical image visualization and processing analysis software, and to write the configuration file and model package; the file processing module is configured to receive the files to be processed from the local or server side, and run the model by calling the processing analysis software; the information processing and display module is configured to monitor the file transfer status and the model operation status in real time in the visualization window console; the server communication module is configured to realize server communication, file transfer and model operation through TCP connection, obtain the files to be processed and send the result files; the image segmentation and analysis module is configured to segment the images stored in the database; the database is configured to store CT image data; the GPU is configured to provide computing power support for image data processing.

[0006] Preferably, the model deployment module includes 3D Slicer, a medical image visualization and processing analysis software, and the file processing module calls 3D Slicer to run the model, encapsulating the entire process to establish a windowed system to control the operation.

[0007] Preferably, the edge acceleration module includes model encapsulation, acceleration dependency enabling and configuration document writing.

[0008] A lung CT image processing method based on an edge-cloud architecture includes: S100 logs in and obtains the server IP address and port; S200 sends commands based on the IP address and port, and establishes a connection after receiving a response; After S300 is successfully connected, the client sends a file request and waits for the server to respond and then transfer the file; After the S400 file to be analyzed is ready, it is introduced into the packaged sub-model and called to run. The running modes are divided into normal mode and AI acceleration card acceleration mode.

[0009] Preferably, the step S300 specifically includes: The S310 client establishes a connection with the TCP 21 / 22 / random port of the server and sends commands; When the S320 client needs to receive data, it sends a PORT command on the channel to read the data; When S330 transmits data, the server connects to the client's specified port through its own TCP port 20 to send data.

[0010] Preferably, the step S320 specifically includes: S321 opens the file receiver and waits for the file data to arrive; S322 has data coming in, reads the data, writes it into a file, and records the number of bytes written; S323 determines the number of bytes written. If the number of bytes written is equal to the file size, the file acceptance is completed. Otherwise, a packet loss command is sent to request retransmission, and S321 is continued.

[0011] Preferably, the step S330 specifically includes: The S331 server starts listening to see if there is a server connection. S332 A client connects, a socket is allocated, and it is ready to receive messages; S333 receives a request from the client to obtain file information and file data, opens a channel and starts transmitting file information and file data.

[0012] A lung CT image processing device based on an edge-cloud architecture comprises a processor and a memory storing program instructions, wherein the processor is configured to execute the lung CT image processing method based on the edge-cloud architecture when running the program instructions.

[0013] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a lung CT image processing method based on an edge-cloud architecture.

[0014] Beneficial effects: The system of the present invention combines edge computing and deep learning technologies to improve the monitoring and treatment of ARDS patients. By implementing deep learning models on edge devices, rapid response and high efficiency are achieved, thereby better meeting the medical field's demand for real-time and accurate data processing. This integrated system is not only expected to improve the treatment level of ARDS patients, but also provide strong support for technological innovation in other related medical fields.

[0015] Instruction Manual Figure 1 is a system architecture diagram of the present invention; Figure 2 Provides a system diagram for model packaging and operation integration; Figure 3 This is the schematic diagram of the communication module; Figure 4 This is the full process flow diagram of the end-to-edge operation of the lung CT image processing system; Figure 5 This is the end-edge-cloud communication flow chart of the lung CT image processing system. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further described in detail below in conjunction with the embodiments and drawings.

[0017] like Figure 1 As shown, a lung CT image processing system based on an edge-cloud architecture includes an edge server, a terminal device, and cloud computing, wherein the terminal device includes: a model deployment module, a file processing module, an information processing and display module, a server communication module, and an image segmentation and analysis module; the edge server includes an edge acceleration module; the cloud computing includes a database and a GPU; the edge server communicates with the terminal device and the cloud computing; the edge acceleration module realizes the rapid operation of the model through an acceleration card; the model deployment module is configured to install medical image visualization and processing and analysis software, and to write configuration files and model packaging; the file processing module is configured to receive files to be processed from a local or server end, and to run the model by calling the processing and analysis software; the information processing and display module is configured to monitor the file transmission status and the model operation status in real time in the visualization window console; the server communication module is configured to realize server communication, file mutual transmission and model operation through a TCP connection, obtain files to be processed and send result files; the image segmentation and analysis module is configured to segment the images stored in the database; the database is configured to store CT image data; the GPU is configured to provide computing power support for image data processing.

[0018] like Figure 2 As shown in the figure, it is a system diagram of model packaging and operation integration. The client and server communication is started through the visual window console, and the files to be processed are transferred to the model's processing folder. Then, the acceleration card startup acceleration function is used to run the model through 3D Slicer to analyze the images of the files to be processed and obtain the result files, which are stored in the result folder.

[0019] like Figure 3 The schematic diagram of the communication module shows that two TCP connections are used between the two communicating hosts. One is a data connection, which is used for data transmission and is full-duplex, allowing simultaneous two-way data transmission. The other is a control connection, which is used to transmit control information (control information such as internal commands and command responses). The client has three components: user interface, client control process, and client data transmission process; the server has two components: server control process and server data transmission process. During the entire interactive FTP session, the control connection is always in a connected state, and the data connection is closed after each file is opened. During this process, the system enables real-time monitoring to observe the connection establishment and file transfer.

[0020] like Figure 4This is the end-to-edge operation flow chart of the lung CT image processing system. The client connects to the server through the server IP address and port, opens the file receiver, and waits for the file to arrive. If there is a file, write the file, record the number of bytes written, and then determine whether the number of bytes written is equal to the file size. This is mainly to determine whether the file sent by the server is complete and whether there is packet loss. If there is packet loss, feedback is given to the server to request resending. If not, the reception is completed. The files to be processed are saved in the model's pending folder, and then the model is started and the analysis results are waited for. After the results are generated, they can be viewed locally or returned to the server.

[0021] like Figure 5 End-edge-cloud communication flow chart of the lung CT image processing system. The server starts the listening port and waits for the client to establish a connection. If a client accesses, a socket is allocated for it and it is ready to receive client messages at any time. When the client has a file request, the server sends the file to the client and waits for a response that the transmission is complete. If the packet is lost, it will be resent.

[0022] Example 1 Next, the specific implementation method of the system operation model analysis data is explained.

[0023] According to the first embodiment provided by the present invention, S1: A lung CT image processing method based on an edge box connected to a PC in the same local area network to call an AI model to analyze data patterns, including the following processes: S2: The edge box first establishes communication with the PC in the LAN, where the edge box acts as the client and the PC acts as the server. There are two ways to establish communication implementation cases: one is to connect through the server console in the integrated system, and the other is to establish a connection and realize file transfer through XFTP.

[0024] S3: Use the first method. After the connection is established, the edge box requests the transfer of the file to be processed. The PC transmits the image file to be analyzed to the edge box client through the console. The client verifies the file format to prevent the file from being mistransmitted and affecting the model operation.

[0025] S4: After the format verification is successful, the client will check whether the file is lost. If the file is lost, the client will respond to the PC server's request for retransmission.

[0026] S5: After the file is transferred, the model is ready to run. Start the run through the console. At this time, the system will run the model script and then call the 3D Slicer software to perform image analysis. After the processing is completed, a PDF format result file will be generated. The result file can be viewed locally or sent to the PC server to complete its request task.

[0027] Example 2 According to the second embodiment provided by the present invention, The lung CT image processing method based on the edge cloud architecture, which communicates with the edge box based on the Inspur cloud server and calls the AI ​​model to analyze the data model, includes the following steps: S1: The edge box sends a connection request to the Inspur cloud server and provides necessary authentication information, such as the unique identifier and key of the edge box.

[0028] S2: After receiving the request, the server performs identity authentication to ensure the legitimacy of the request. Once the identity authentication is successful, the two parties begin to negotiate the communication protocol, including the determination of data transmission protocol, encryption method, etc.

[0029] S3: After the connection is established, the edge box requests the transfer of the file to be processed through the established communication channel. After receiving the request, the cloud server transmits the image file to be analyzed to the edge box client through the console or other means. This process needs to consider the file size, transmission speed and network stability to ensure the integrity and accuracy of the file.

[0030] S4: After the file is transmitted, the edge box client performs a format check on the file to ensure the correctness of the file. At the same time, the system also checks whether there is packet loss in the file. If there is packet loss, the client will send a request to the cloud server for retransmission to ensure the complete transmission of the data.

[0031] S5: After the file transfer is complete, the system is ready to run the model. Start the run through the console, the system executes the model script, and calls the 3D Slicer software for image analysis. This step is the core of the entire process. Through the efficient operation and accurate analysis of the model, the system can handle complex data tasks.

[0032] S6: After the image analysis is completed, the system generates a result file in PDF format. The user can choose to view the results locally or send the results to the PC server.

[0033] It should be noted that this design takes into account the flexible needs of users, so that the processed data can be easily viewed locally or in the cloud. Through this carefully designed communication process, efficient and reliable communication between Inspur cloud servers and edge boxes is ensured, providing a solid foundation for model operation and data analysis. This communication mechanism not only takes security and stability into consideration, but also introduces a checksum check mechanism in the file transfer stage to deal with possible transmission errors. Such a system design helps to improve the stability and availability of the overall system.

[0034] The software part has two parts, one is the local integrated system, the other is the server file transfer system. The local integrated system framework consists of two Headers files, three Sources files and one Forms file. The Headers files are mainwindow.h and sendfile.h, the Sources files are main.cpp, mainwindow.cpp and sendfile.cpp, and the Forms file is mainwindow.ui. In the mainwindow.cpp file, you can see that #include "ui_mainwindow.h" references such a header file. After opening it, you can find QObject::connect(ExitPushBotton, SIGNAL(clicked()), MainWindow, SLOT(close())); The underlying principle is to connect the signal of clicking the Exit button to the slot function of closing the interface. The signal is emitted by the button PushBotton and received by the main window MainWindow. To understand more simply: button: signal sender->ExitPushBotton; window: signal receiver and processor->MainWindow; signal sent: clicked->clicked; slot function: how to handle->close. The Sources file mainly contains the functions of each module, among which the QT components used are: QLabel, QLineEdit, QPushButton, QProgressBar, QTextBrowser and QTextEdit.

[0035] The system is packaged as an executable file and placed on the desktop. Users only need to click to run it. In the local operation mode in the main interface, the QPushButton button is used to select the model image file to be analyzed, and the slot function SelectFile() is established, and then it is sent to the model specified folder to be processed. Here, if the target folder is not empty, it needs to be emptied before it can be stored. The slot function UploadFile is established to achieve this. The QDir class needs to be introduced, which provides a system-independent way to access the system directory structure and its contents. QLineEdit is used to record and display the path of the selected file, and QProgressBar is used to monitor the progress of the record file upload. Here, QTextBrowser is used to monitor the progress of the Terminal side when the printing model is running (only for the printing of the model running status). A pseudo terminal needs to be established, but this solution is too complicated. The simplest method is adopted, that is, using QT to start a Terminal command and capture its output. Here, the activities in the terminal window that the user has opened will not be captured, but the output of the specific command started by this QT program can be captured, such as: terminalProgress->start("bash -c 'ls -a ~'"). QTextEdit is used to record changes in PDF files in the monitored folder and interactively prompt users through QMessageBox. Here, a member variable of QFileSystemWatcher is added to the main window class, initialized in the constructor and connected to the slot function to implement its function.

[0036] The main window also provides a server connection module, in which an object of the SendFile class is created by using the constructor inherited from the QObjectSendFile class. The ConnectServer function is used to connect to the server. It creates a QTcpSocket object m_tcp and uses connectToHost to connect to the specified IP address and port. After the connection is successful, the ConnectOK signal is triggered; when the connection is disconnected, the Over signal is triggered. After the connection is successful, the object remains connected. The SendFileToServer function is used to send files to the server. It opens the file of the specified path, obtains the file size, and sends the file content line by line to the server through the QTcpSocket object. When the file is sent for the first time, the file size information is sent. During the sending process, the CurPercent signal is triggered to indicate the progress of the transmission. In the constructor of the main window function, a thread object and a task object are created, the task object is moved to the thread for execution, and a series of signals and slots are connected. This ensures that the file transfer and connection operations are performed in independent threads to prevent interface jams. On the receiving file side, a class for receiving files through TCP sockets is used. The constructor of the RecvFile class is defined, which accepts a pointer of QTcpSocket type as a parameter and saves it in the private member m_socket of the class. Here, the run function is a thread execution function rewritten from QThread. In this function, a file named recv.txt is created and opened in write mode. After that, a Lambda function is triggered when there is data to read by connecting the readyRead signal of m_socket. In the Lambda function, first check whether count is 0. If so, 4 bytes of data are read through m_socket->read((char*)&total, 4);, which represents the total size of the file. Next, the remaining data is read through m_socket->readAll() and written to the file. At the same time, count is updated to track the number of bytes received. Finally, when count reaches total, it means that the file is received. Close the socket and file, and then emit the Over signal.

[0037] To enable the domestic AI acceleration card, first install the driver as required. Then when accelerating the model, you need to execute the corresponding commands for compilation, write the configuration file, and then write the execution command into a script for the visualization window system to call.

[0038] The present invention constructs a lung CT image processing system based on the edge-cloud architecture, then encapsulates the AI ​​model into a visual software, forms an integrated operating system with the edge box, and uses the acceleration card carried by the edge box to realize rapid analysis of the model. Among them, a TCP connection is also established to realize file transmission between the local end and the server end. The transmitted files are mainly image files to be analyzed and processed and analysis result files. The portable integrated system formed by combining the above functions and structures is simple to use and fast to analyze. It can realize high-speed analysis of CT images of ARDS patients, quickly locate the cause of the disease, and provide diagnosis and treatment plans.

Claims

1. A lung CT image processing system based on edge-cloud architecture, characterized in that: It includes an edge server, a terminal device, and cloud computing. The terminal device includes: a model deployment module, a file processing module, an information processing and display module, a server communication module, and an image segmentation and analysis module; the edge server includes an edge acceleration module; the cloud computing includes a database and a GPU; the edge server communicates with the terminal device and the cloud computing; The edge acceleration module realizes the rapid operation of the model through the acceleration card; the model deployment module is configured to install the medical image visualization and processing analysis software, and to write the configuration file and model package; the file processing module is configured to receive the files to be processed from the local or server side, and run the model by calling the processing analysis software; the information processing and display module is configured to monitor the file transfer status and the model operation status in real time in the visualization window console; the server communication module is configured to realize server communication, file transfer and model operation through TCP connection, obtain the files to be processed and send the result files; the image segmentation and analysis module is configured to segment the images stored in the database; the database is configured to store CT image data; the GPU is configured to provide computing power support for image data processing.

2. The lung CT image processing system based on edge-cloud architecture according to claim 1, characterized in that: The model deployment module includes the medical image visualization and processing analysis software 3D Slicer, and the file processing module runs the model by calling 3DSlicer, encapsulating the entire process and establishing a window system to control the operation.

3. The lung CT image processing system based on edge-cloud architecture according to claim 1, characterized in that: The edge acceleration module includes model encapsulation, acceleration dependency enabling and configuration document writing.

4. A lung CT image processing method based on an edge-cloud architecture using the system of any one of claims 1 to 3, characterized in that: include: S100 logs in and obtains the server IP address and port; S200 sends commands based on the IP address and port, and establishes a connection after receiving a response; After S300 is successfully connected, the client sends a file request and waits for the server to respond and then transfer the file; After the S400 file to be analyzed is ready, it is introduced into the packaged sub-model and called to run. The running modes are divided into normal mode and AI acceleration card acceleration mode.

5. The lung CT image processing method based on the edge-cloud architecture according to claim 4 is characterized in that: The step S300 specifically includes: The S310 client establishes a connection with the TCP 21 / 22 / random port of the server and sends commands; When the S320 client needs to receive data, it sends a PORT command on the channel to read the data; When S330 transmits data, the server connects to the client's specified port through its own TCP port 20 to send data.

6. The lung CT image processing method based on the edge-cloud architecture according to claim 5 is characterized in that: The step S320 specifically includes: S321 opens the file receiver and waits for the file data to arrive; S322 has data coming in, reads the data, writes it into a file, and records the number of bytes written; S323 determines the number of bytes written. If the number of bytes written is equal to the file size, the file acceptance is completed. Otherwise, a packet loss command is sent to request retransmission, and S321 is continued.

7. The lung CT image processing method based on the edge-cloud architecture according to claim 5 is characterized in that: The step S330 specifically includes: The S331 server starts listening to see if there is a server connection. S332 A client connects, a socket is allocated, and it is ready to receive messages; S333 receives a request from the client to obtain file information and file data, opens a channel and starts transmitting file information and file data.

8. A lung CT image processing device based on an edge-cloud architecture, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the lung CT image processing method based on the edge-cloud architecture as described in any one of claims 4-7 when running the program instructions.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the method described in any one of claims 4 to 7 is implemented.

Citation Information

Patent Citations

  • Medical service system

    CN110473623A

  • Portable CT (Computed Tomography) visualization equipment

    CN114266856A

  • Edge computing system based on automatic federated learning and learning method thereof

    CN114638357A

  • Business management method and system, configuration server and edge computing device

    CN114979246A

  • Hybrid quantization of neural networks for edge computing applications

    US20220044114A1