SPECT data processing system and method based on NVIDIA AGX ORIN platform

Abstract

The invention provides an SPECT data processing system and method based on an NVIDIA AGX ORIN platform, and the system is characterized in that a detector is connected with the input end of a digital-to-analog conversion board, and the output end of the digital-to-analog conversion board is connected with the input end of a data transfer board card; the AGX data collection board is respectively connected with the data transfer board card and the upper computer; the detector is used for collecting analog quantity data and transmitting the collected analog quantity data to the digital-to-analog conversion plate; the digital-to-analog conversion board is used for converting the received analog quantity data into digital signals and transmitting the digital signals to the data transfer board card; the data transfer board card is used for receiving the digital signal and transmitting the digital signal to the AGX data collection board; the AGX data collection board is used for receiving the digital signal, processing the digital signal to generate image data, and transmitting the generated image data to the upper computer; and the upper computer is used for receiving and displaying the image data.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a SPECT data processing system and method based on the NVIDIA AGX ORIN platform. Background Art

[0002] SPECT imaging technology provides rich image information for nuclear medicine diagnosis through SPECT functional imaging. However, with the continuous development of imaging equipment, the amount of data is increasing, and the imaging technology is becoming more and more accurate, so a larger storage space must be used, and a faster processing speed is required to improve equipment performance more quickly and accurately and improve diagnostic efficiency.

[0003] As attached Figure 2 As shown in the figure, the traditional SPECT data processing system consists of a host computer, an FPGA data collection board with ARM, a data conversion board and a detector. Among them, data analysis and image fusion are all implemented by the host computer, resulting in a large amount of unstable data transmission in the system. It is necessary to rely on network ports, network cables, switches, etc. to ensure the stability of data transmission. At the same time, the traditional SPECT data processing system has the problems of slow calculation speed and low processing accuracy. When the SPECT detector generates a large amount of data and faces complex computing requirements, it cannot meet the requirements of real-time diagnosis and rapid decision-making in clinical applications.

[0004] Patent document CN116671946B (application number: 202310405711.6) discloses a method for reconstructing dynamic images based on SPECT dynamic acquisition data. This method simplifies the problem of solving the value of all voxels in conventional tomographic image reconstruction and the problem of solving the weighted coefficients of the influence of different organs and tissues. By reducing the number of unknown variables in the image reconstruction process, the sampling angle and noise level requirements of the projection data required for the reconstruction of tomographic data at each time point are reduced, thereby being able to more accurately reconstruct dynamic tomographic images based on the image distribution of different organs, which is convenient for subsequent image quantitative analysis and clinical diagnosis. Summary of the invention

[0005] In view of the defects in the prior art, the object of the present invention is to provide a SPECT data processing system and method based on the NVIDIA AGX ORIN platform.

[0006] A SPECT data processing system based on the NVIDIA AGX ORIN platform provided by the present invention includes: a host computer, an AGX data collection board, a data transfer board, a digital-to-analog conversion board, and a detector;

[0007] The detector is connected to the input end of the digital-to-analog conversion board, and the output end of the digital-to-analog conversion board is connected to the input end of the data transfer board; the AGX data collection board is connected to the data transfer board and the host computer respectively;

[0008] The detector is used to collect analog data and transmit the collected analog data to the digital-to-analog conversion board;

[0009] The digital-to-analog conversion board is used to convert the received analog data into digital signals, and transmit the digital signals to the data transfer board;

[0010] The data transfer board is used to receive digital signals and transmit the digital signals to the AGX data collection board;

[0011] The AGX data collection board is used to receive digital signals, process the digital signals to generate image data, and transmit the generated image data to the host computer;

[0012] The host computer is used to receive image data and display it.

[0013] Preferably, the data transfer board includes an FPGA chip;

[0014] The FPGA chip includes multiple output pins; the multiple output pins include: 40 GPIO pins, 48 ​​PCIE pins, 32 MIPI CSI pins, 68 M.2SATA protocol pins and 75 M.2NVMe SSD protocol pins;

[0015] Based on the 40 GPIO pins, the 48 PCIE pins, the 32 MIPI CSI pins, the 68 M.2SATA protocol pins and the 75 M.2NVMe SSD protocol pins, communication between the FPGA chip in the data transfer board and the NVIDIA AGX ORIN chip in the AGX data collection board is achieved.

[0016] Preferably, the AGX data collection board includes a PCIEX16 interface and an eSATA interface;

[0017] The PCIEX16 interface is used to process data with a usage frequency less than or equal to a preset value based on cold storage;

[0018] The eSATA interface is used to process data with a usage frequency greater than a preset value based on thermal storage.

[0019] Preferably, the AGX data collection board includes an NVIDIA AGX ORIN chip;

[0020] The NVIDIA AGX ORIN chip is used to convert the collected digital signals into image data.

[0021] According to a SPECT data processing method based on the NVIDIA AGX ORIN platform provided by the present invention, the following steps are performed using the SPECT data processing system based on the NVIDIA AGX ORIN platform described above:

[0022] Step S1: using the detector to collect analog data, and transmitting the collected analog data to the digital-to-analog conversion board;

[0023] Step S2: The digital-to-analog conversion board converts the received analog data into digital signals, and transmits the digital signals to the data transfer board;

[0024] Step S3: The data transfer board receives the digital signal and transmits the digital signal to the AGX data collection board;

[0025] Step S4: the AGX data collection board receives the digital signal, processes the digital signal to generate image data, and transmits the generated image data to the host computer;

[0026] Step S5: The host computer receives the image data and displays it.

[0027] Preferably, the step S4 comprises:

[0028] Step S4.1: preprocessing the received digital information to obtain preprocessed digital information;

[0029] Step S4.2: Process the pre-processed digital information to obtain image data;

[0030] Step S4.3: preprocessing the generated image data to obtain preprocessed image data;

[0031] Step S4.4: Transmit the preprocessed image data to the host computer.

[0032] Preferably, the step S4.1 comprises: performing preprocessing on the received digital information including removing noise and outliers to obtain preprocessed digital information.

[0033] Preferably, the step S4.3 comprises: optimizing the generated image data according to a deep learning algorithm;

[0034] The optimization of the generated image data according to the deep learning network includes: using the deep learning network to perform noise reduction, contrast enhancement, pixel value reconstruction of missing areas, image segmentation, lesion area identification and position marking on the generated images to obtain optimized image data.

[0035] Preferably, the AGX data collection board realizes real-time streaming processing of image data based on the multi-task parallel processing capability of the NVIDIA AGX ORIN chip.

[0036] Preferably, the data transfer board adds a frame header, version number, data length, data load, frame tail and checksum to the output data; when the AGX data collection board receives a digital signal, it completes the confirmation of the transmitted data including the frame header, version number, data length, data load and frame tail to obtain a checksum. When the currently obtained checksum is consistent with the checksum added by the data transfer board, the data verification is successful and the data is not lost or damaged.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. NVIDIA AGX ORIN has extremely high parallel computing and artificial intelligence processing capabilities. When applied to SPECT data processing systems, it can accelerate the processing of medical imaging data, optimize imaging quality, and reduce data transmission requirements.

[0039] 2. The reintegrated cables of the data transfer board are used. The data transfer board has multiple output pins for connecting with the AGX data collection board, which solves the problem of insufficient LVDS protocol IO pin resources of the AGX data collection board.

[0040] 3. Use NVIDIA AGX ORIN's deep learning model to reconstruct missing or incomplete images, and effectively repair the black edges and gaps that appear during SPECT imaging.

[0041] 4. Configure the PCIEX16 interface and eSATA interface on the AGX data collection board, use hot storage (such as SSD) to process the data with high current usage frequency, and use cold storage (such as large HDD or cloud storage) for infrequently used data (such as historical images), which can effectively improve data storage efficiency.

[0042] 5. By adding a unique frame header, version number, data length and data payload to the output data, and adding a checksum and frame tail to the input of the AGX data collection board, the integrity of the data can be confirmed, and when imaging problems occur, the problem location can be quickly located.

[0043] 6. The present invention provides a SPECT data processing system based on the NVIDIA AGX ORIN platform to improve image processing speed and quality and optimize clinical workflow. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0045] Figure 1 Schematic diagram of the topology of the SPECT data processing system based on the NVIDIA AGX ORIN platform.

[0046] Figure 2 Schematic diagram of the topological structure of the traditional SPECT data processing system.

[0047] Figure 3 Flowchart of image processing for AGX.

[0048] Figure 4 The image before processing has fusion gaps.

[0049] Figure 5 This is a schematic diagram of the connection relationship between the data transfer board and the AGX data collection board.

[0050] Figure 6 This is a schematic diagram of the connection relationship between the data transfer board and the AGX data collection board. DETAILED DESCRIPTION

[0051] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0052] Example 1

[0053] A SPECT data processing system based on the NVIDIA AGX ORIN platform provided by the present invention includes: a host computer, an AGX data collection board, a data transfer board, a digital-to-analog conversion board, and a detector;

[0054] The detector is connected to the input end of the digital-to-analog conversion board, and the output end of the digital-to-analog conversion board is connected to the input end of the data transfer board; the AGX data collection board is connected to the data transfer board and the host computer respectively;

[0055] The detector is used to collect analog data and transmit the collected analog data to the digital-to-analog conversion board;

[0056] The digital-to-analog conversion board is used to convert the received analog data into digital signals, and transmit the digital signals to the data transfer board;

[0057] The data transfer board is used to receive digital signals and transmit the digital signals to the AGX data collection board;

[0058] The AGX data collection board is used to receive digital signals, process the digital signals to generate image data, and transmit the generated image data to the host computer;

[0059] The host computer is used to receive image data and display it.

[0060] Specifically, the data transfer board includes an FPGA chip;

[0061] Since the NVIDIA AGX ORIN chip itself is packaged with the above-mentioned pins: 40 GPIO pins, 48 ​​PCIE pins, etc., the FPGA chip in the data transfer board can only convert the pin protocol in order to match the fixed package pins of the NVIDIA AGX ORIN chip; therefore, the FPGA chip includes a variety of output pins; the multiple output pins include: 40 GPIO pins, 48 ​​PCIE pins, 32 MIPI CSI pins, 68 M.2SATA protocol pins and 75 M.2NVMe SSD protocol pins;

[0062] Only based on the 40 GPIO pins, the 48 PCIE pins, the 32 MIPI CSI pins, the 68 M.2SATA protocol pins and the 75 M.2NVMe SSD protocol pins can the communication between the FPGA chip in the data transfer board and the NVIDIA AGX ORIN chip in the AGX data collection board be realized.

[0063] Specifically, the AGX data collection board includes a PCIEX16 interface and an eSATA interface;

[0064] The PCIEX16 interface is used to process data with a usage frequency less than or equal to a preset value based on cold storage;

[0065] The eSATA interface is used to process data with a usage frequency greater than a preset value based on thermal storage.

[0066] Specifically, the AGX data collection board includes an NVIDIA AGX ORIN chip;

[0067] The NVIDIA AGX ORIN chip is used to convert the collected digital signals into image data.

[0068] According to a SPECT data processing method based on the NVIDIA AGX ORIN platform provided by the present invention, the following steps are performed using the SPECT data processing system based on the NVIDIA AGX ORIN platform described above:

[0069] Step S1: using the detector to collect analog data, and transmitting the collected analog data to the digital-to-analog conversion board;

[0070] Step S2: The digital-to-analog conversion board converts the received analog data into digital signals, and transmits the digital signals to the data transfer board;

[0071] Step S3: The data transfer board receives the digital signal and transmits the digital signal to the AGX data collection board;

[0072] Step S4: the AGX data collection board receives the digital signal, processes the digital signal to generate image data, and transmits the generated image data to the host computer;

[0073] Step S5: The host computer receives the image data and displays it.

[0074] Specifically, step S4 includes:

[0075] Step S4.1: preprocessing the received digital information to obtain preprocessed digital information;

[0076] Step S4.2: Process the pre-processed digital information to obtain image data;

[0077] Step S4.3: preprocessing the generated image data to obtain preprocessed image data;

[0078] Step S4.4: Transmit the preprocessed image data to the host computer.

[0079] Specifically, the step S4.1 includes: performing preprocessing on the received digital information including removing noise and abnormal values ​​to obtain preprocessed digital information.

[0080] Specifically, the step S4.3 includes: optimizing the generated image data according to the deep learning algorithm;

[0081] The optimization of the generated image data according to the deep learning network includes: using the deep learning network to perform noise reduction, contrast enhancement, pixel value reconstruction of missing areas, image segmentation, lesion area identification and position marking on the generated images to obtain optimized image data.

[0082] Specifically, the AGX data collection board realizes real-time streaming processing of image data based on the multi-task parallel processing capability of the NVIDIA AGX ORIN chip.

[0083] Specifically, the data transfer board adds a frame header, version number, data length, data load, frame tail and checksum to the output data; when the AGX data collection board receives a digital signal, it completes the confirmation of the transmitted data including the frame header, version number, data length, data load and frame tail to obtain a checksum. When the currently obtained checksum is consistent with the checksum added by the data transfer board, the data verification is successful and the data is not lost or damaged.

[0084] Example 2

[0085] Embodiment 2 is a preferred embodiment of Embodiment 1

[0086] like Figure 1 As shown, a SPECT data processing system based on the NVIDIA AGX ORIN platform provided by the present invention includes: a host computer, an AGX data collection board, a data transfer board, a digital-to-analog conversion board and a detector;

[0087] The input end of the digital-to-analog conversion board is connected to the detector, the output end of the digital-to-analog conversion board is connected to the input end of the data transfer board, and the AGX data collection board is connected to the data transfer board and the host computer respectively. The digital-to-analog conversion board is used to convert the analog signal collected by the detector into a digital signal and output it to the data transfer board; the data transfer board is used to receive the digital signal output by the digital-to-analog conversion board; the detector is used to collect analog data; the AGX data collection board is used to process the data output by the data transfer board and generate image data.

[0088] The data transfer board includes an FPGA chip, and the FPGA chip includes multiple output pins. The multiple output pins include 40 GPIO pins, 48 ​​PCIE pins, 32 MIPI CSI pins, 68 M.2SATA protocol pins, and 75 M.2NVMe SSD protocol pins.

[0089] The AGX data collection board can directly perform data analysis, image fusion and imaging on the data output by the data transfer board, and can directly output the imaging work that needs to be completed by the host computer through wireless transmission. Due to the stability of the image format transmission process, the traditional working mode eliminates the need to use an FPGA chip with an ARM core in the data collection board, and then transmit the acquired data through the PHY protocol; the physical layer reduces network ports, network cables, switches, etc., thereby reducing the total amount of data transmission and enhancing the stability of data transmission.

[0090] The AGX data collection board includes an NVIDIA AGX ORIN chip;

[0091] The NVIDIA AGX ORIN chip optimizes image data based on deep learning algorithms, handles image cutting problems, increases image contrast, reduces image noise, and uses AI analysis to automatically highlight lesions and indicate their locations. In this way, a highly complete image can be generated in the internal storage module of the AGX data collection board. Since it is image data, the transmission is extremely stable and can be transmitted using a wireless transmission module, without the need for traditional fiber optic transmission and network port transmission (fiber optics are easily damaged and network port transmission data is not stable enough).

[0092] like Figure 3 As shown, the image processing process of the AGX data collection board includes the following steps:

[0093] The image processing process of the AGX data collection board is all completed based on the NVIDIA AGX ORIN chip.

[0094] The image processing process of the AGX data collection board is that in SPECT imaging, the scanning probe rotates around the target and collects the gamma-ray signals emitted by the radioisotope from multiple angles, thus obtaining multi-view two-dimensional projection image data with position information. The data is transferred to the AGX data collection board through the data transfer board. The AGX data collection board is based on the NVIDIA AGX ORIN chip, and its function is to reconstruct and optimize the multi-view two-dimensional projection image data with position information into three-dimensional functional images and anatomical structure images.

[0095] Data acquisition: Start the SPECT system to collect data and collect the patient's medical imaging data.

[0096] Data preprocessing: Perform preliminary processing on the collected data to remove noise and outliers, ensure data quality, and lay the foundation for subsequent analysis.

[0097] GPU parallel processing: Leverage the powerful computing power of the NVIDIA AGX ORIN chip for efficient data processing.

[0098] CUDA acceleration: CUDA technology is used to realize parallel computing of image reconstruction and processing algorithms, thereby significantly improving processing speed and performance.

[0099] Fast Iteration Algorithm: Continuously improve the reconstruction results based on image acquisition data, gradually optimize image quality and speed up processing. By implementing fast matrix operations on the GPU, this iterative process can provide real-time feedback and continuously adjust parameters to further improve the clarity and accuracy of the reconstructed image.

[0100] AI Noise Reduction: Applies a GPU-based denoising algorithm to reduce artifacts and interfering noise in images.

[0101] Contrast Enhancement: Increases image contrast to make details clearer and improve visibility.

[0102] AI analysis: Use deep learning and machine learning technologies to perform intelligent analysis on processed images.

[0103] Automatic lesion detection: Automatically identify possible lesions in images through trained AI models, thereby improving the efficiency and accuracy of diagnosis.

[0104] Dynamic Artifact Correction: Dynamically adjust algorithm parameters according to different image features to enhance detection sensitivity and accuracy.

[0105] Output optimized images: Finally, optimized and analyzed images are generated for clinical diagnosis and further evaluation.

[0106] Specifically, the image processing of the AGX data collection board has the following advantages:

[0107] Optimize image quality: Use deep learning models to accelerate traditional iterative reconstruction algorithms, such as algebraic iterative reconstruction (ART) or maximum likelihood expectation maximization (MLEM) algorithms, and use deep learning to predict the initial solution to reduce the number of iterations, improve reconstruction speed and image quality. With the high parallelism of NVIDIA AGX ORIN's GPU architecture, the intensive computing tasks required in the reconstruction process are perfectly parallelized, thereby accelerating the calculation.

[0108] Specifically, the spatial resolution of the image is improved by generating an adversarial network (GAN). The generative network and the discriminative network are trained adversarially to generate output images with higher resolution than the input while retaining important details and realism.

[0109] Use convolutional neural networks to upsample images and compensate for details, improving the quality of each pixel by learning patterns from a large number of data samples.

[0110] The denoising autoencoder is used to learn to remove various noises introduced during the SPECT imaging process. Through the automatic learning of the network, the denoising autoencoder can effectively remove noise points under low signal-to-noise ratio conditions after long-term training.

[0111] Deploy deep neural networks such as the U-Net structure to reduce artifacts during imaging. Define a small sample training model using artifacts in the training set to improve the system's ability to adaptively remove artifacts under different conditions.

[0112] By inputting the SPECT data into the deep fusion neural network separately, the dynamic range and contrast of the image are improved. The network learns how to combine the two modalities to achieve the best image fusion effect.

[0113] Use deep learning techniques (such as U-Net) to achieve accurate organ segmentation and lesion area identification, ensuring automatic identification of key features on high-resolution images.

[0114] Reduce processing time:

[0115] Real-time data analysis and processing

[0116] Data stream optimization: Utilizing the multitasking capabilities of NVIDIA AGX ORIN to achieve real-time streaming processing of image data can significantly reduce the storage and reading time of the entire image data.

[0117] NVIDIA AGX ORIN is equipped with a powerful GPU capable of massive parallel computing, which means that multiple image pixels can be processed simultaneously, greatly speeding up image reconstruction and processing.

[0118] NVIDIA AGX ORIN provides powerful edge computing capabilities, enabling real-time analysis and processing at the data collection site, thereby reducing data transmission delays. By implementing a fast iterative reconstruction algorithm, the digital signals collected by SPECT can be instantly reconstructed to ensure that imaging data is generated in the shortest possible time.

[0119] Using trained deep learning models to optimize image quality, such as noise reduction and contrast enhancement, helps improve image clarity, allowing higher-quality diagnostic images to be obtained in the same amount of time. Intelligent algorithms can automatically identify and correct artifacts caused by patient movement, further improving image quality and processing efficiency.

[0120] Through real-time image analysis, NVIDIA AGX ORIN has the ability to provide doctors with instant diagnostic prompts, significantly reducing doctors' analysis and judgment time. The system can automatically generate diagnostic reports based on image analysis, further improving work efficiency and shortening patient waiting time.

[0121] Reduced data transmission requirements: NVIDIA AGX ORIN's edge computing capabilities allow a large amount of preliminary data processing to be completed locally, reducing dependence on high-bandwidth connections and the need to transfer large amounts of raw data to and from servers. When the network environment is poor or even disconnected, NVIDIA AGX ORIN can still perform its critical processing tasks without relying on a continuous network connection, which is a major advantage over traditional host computer systems. Since most data processing tasks are performed locally, this reduces the number of times sensitive medical data is transmitted over the network, reducing potential data exposure and leakage risks. Less data transmission also means less privacy risk, thereby complying with medical data protection regulations.

[0122] like Figure 1 As shown in the figure, the data transmitted by a detector is only an analog signal with position information obtained by one acquisition module. If you want to achieve the usable image standard, it needs to be converted into a digital signal for use. One ADC data conversion board can upload 8 acquisition modules, which means that the output digital signal is a differential signal pre-processed by the amplified signal. The differential signal pre-processed by the ADC conversion chip must be output through the FPGA chip of the ADC conversion board. At least 4 LVDS differential signals must be input to the data transfer board. 4 data transfer boards need to process 32 ADC conversion chip boards. Each data transfer board needs to input 4 LVDS differential lines (X2) multiplied by 8 boards. Therefore, each data transfer board must reserve at least 64 LVDS input pins and 64 LVDS output pins. The four data transfer boards output a total of 256 LVDS pins to the IO pins of the AGX data acquisition and processing board. However, in the back-end NVIDIA AGX The pin resources of the ORIN main control chip are only 40 GPIO pins, which are not enough to meet the pin resources of the signal acquisition data at the input end. Therefore, the present invention designs an FPGA chip in the data transfer board to output 40 GPIO pins, 48 ​​PCIE pins, 32 MIPI CSI pins, 68 M.2SATA protocol pins, and 75 M.2NVMe SSD protocol pins. Figure 5-6 shown.

[0123] Although AGX has advantages such as optimizing image quality, there are problems in hardware connection. Because the NVIDIA AGXORIN module has fixed configurations such as PCIE protocol, USB protocol, MIPI protocol and other fixed protocols, when the ADC outputs a differential data signal of the LVDS protocol, the AGX data collection board cannot interact with the ADC data conversion board. Therefore, the present invention installs an FPGA chip on the data transfer board to perform data transfer and data conversion. The specific operation steps are as follows:

[0124] FPGA directly calls the 7series integrated Block for PCl Express (3.3) IP core to configure the IP core;

[0125] The BAR0 and BAR1 areas exchange data, and the PCIE protocol data interaction is completed by configuring the IP core's clock, reset signal, input data, input data enable and other valid information through the program.

[0126] The specific pin correspondence is as follows: the hardware connection uses a soft PCB cable, and the configured BANK10 pins are directly connected to the PCIEX16 pins on the NVIDIA AGX ORIN side, and the 48 differential PIN pins are interconnected:

[0127] The MIPI CSI protocol pins, M.2SATA protocol pins, and M.2NVMe SSD protocol pins are the same as above. They are all compatible with the fixed pins of the existing AGX modules by configuring the FPGA IP core address, so that 256 data lines can be uploaded to the AGXOrin module, and the NVIDIA AGX ORIN module performs image fusion and optimization.

[0128] There are as many as 256 data lines between the data transfer board and the AGX data acquisition board. During the data transmission process, data conflicts or losses may occur due to multi-source data collection, or data may be damaged due to unexpected machine crashes. Therefore, it is very important to set the receiving format at the receiving end. During the data acquisition and transmission process, a unique frame header, version number, data length, and data load are added to each data signal line through the FPGA chip on the data transfer board, and a checksum and frame tail are added at the input end of the AGX data collection board.

[0129] First, you need to define the specific structure of the data frame.

[0130] Frame header: For example, a fixed 16-bit value (such as 0x3C3C) is used.

[0131] Data Length: For example, 2 bytes, indicating the number of bytes of the data payload.

[0132] Data Payload: variable length, the specific content depends on the actual data.

[0133] Checksum: 2 bytes, using CRC32 to check data integrity.

[0134] Frame tail (Tail): For example, a fixed 16-bit value (such as 0xFCFC) is used.

[0135] After receiving the data, you need to parse and verify the frame structure to confirm the frame header, version number, data length, data payload, and checksum.

[0136] This ensures data integrity and allows quick location of any imaging issues.

[0137] like Figure 4 As shown, since the final imaging of SPECT is composed of four probes, each probe has 64 detector crystal acquisition modules. The edges of the 64 modules have obvious black edges due to the physical principles of the crystals, resulting in 7X7 horizontal and vertical black edges in the image, which makes the image have obvious gaps and increases the difficulty of reading for medical staff. The deep learning model of AGX is used to reconstruct missing or incomplete images. For example, a convolutional neural network (CNN) is used to predict the pixel values ​​of the missing area to generate a complete image. Deep learning is used for image restoration, and the information of the gap area in the image is input to generate the missing part and smoothly transition to the surrounding pixels.

[0138] First, through the noise algorithm, filtering such as Gaussian filtering and median filtering is performed in the NVIDIA AGX ORIN module to reduce random noise in the image. Then, edge enhancement filters (such as Laplacian filter, Sobel filter, etc.) are used to highlight the image edges to help improve the visual effect of the image. After acquiring the image, for the black edges of the SPECT image, local interpolation or image reconstruction technology is used to try to fill the black edge area. Then, the pixel values ​​of the missing area are predicted by using a convolutional neural network (CNN) to generate a complete image. Image restoration is performed using deep learning, and the information of the gap area in the image is input to generate the missing part and smoothly transition to the surrounding pixels. Convolutional Neural Networks (CNN) is a type of deep learning model that is widely used in image processing, computer vision and other related fields. CNN can effectively extract and learn spatial hierarchical features in images by simulating the structure of biological visual systems. Custom image processing solutions can also be implemented by using NVIDIA's deep learning frameworks (such as PyTorch or TensorFlow) and image processing libraries (such as OpenCV).

[0139] This is difficult for the host computer to do.

[0140] SPECT produces a large number of high-resolution medical images, which need to be completely preserved during diagnosis and analysis, including patient information, acquisition parameters, equipment information, image acquisition time, etc., as well as real-time data obtained from the equipment, test results, and subsequently generated report data. Medical images are generally required to be stored for many years to meet clinical needs and legal requirements, so it is necessary to calculate the demand for long-term storage space. For different types of data, hot storage (such as SSD) can be used to process data that is currently used more frequently, while cold storage (such as large HDD or cloud storage) is used for infrequently used data (such as historical images). The present invention uses PCIE resource pins that are not used by NVIDIA AGX ORIN to design a PCIEX16 interface, which is convenient for using high-performance SSDs (solid state drives) to process data that is currently used more frequently, and designs an eSATA interface for using HDD hard disk drives to store infrequently used historical data, which can effectively improve data storage efficiency.

[0141] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0142] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A SPECT data processing system based on the NVIDIA AGX ORIN platform, characterized in that: include: Host computer, AGX data collection board, data transfer board, digital-to-analog conversion board and detector; The detector is connected to the input end of the digital-to-analog conversion board, and the output end of the digital-to-analog conversion board is connected to the input end of the data transfer board; The AGX data collection board is connected to the data transfer board and the host computer respectively; The detector is used to collect analog data and transmit the collected analog data to the digital-to-analog conversion board; The digital-to-analog conversion board is used to convert the received analog data into digital signals, and transmit the digital signals to the data transfer board; The data transfer board is used to receive digital signals and transmit the digital signals to the AGX data collection board; The AGX data collection board is used to receive digital signals, process the digital signals to generate image data, and transmit the generated image data to the host computer; The host computer is used to receive image data and display it.

2. The SPECT data processing system based on the NVIDIA AGX ORIN platform according to claim 1, characterized in that: The data transfer board includes an FPGA chip; The FPGA chip includes multiple output pins; The multiple output pins include: 40 GPIO pins, 48 ​​PCIE pins, 32 MIPI CSI pins, 68 M.2SATA protocol pins and 75 M.2NVMe SSD protocol pins; Based on the 40 GPIO pins, the 48 PCIE pins, the 32 MIPI CSI pins, the 68 M.2SATA protocol pins and the 75 M.2NVMe SSD protocol pins, communication between the FPGA chip in the data transfer board and the NVIDIA AGX ORIN chip in the AGX data collection board is achieved.

3. The SPECT data processing system based on the NVIDIA AGX ORIN platform according to claim 1, characterized in that: The AGX data collection board includes a PCIEX16 interface and an eSATA interface; The PCIEX16 interface is used to process data with a usage frequency less than or equal to a preset value based on cold storage; The eSATA interface is used to process data with a usage frequency greater than a preset value based on thermal storage.

4. The SPECT data processing system based on the NVIDIA AGX ORIN platform according to claim 1, characterized in that: The AGX data collection board includes an NVIDIA AGX ORIN chip; The NVIDIA AGX ORIN chip is used to convert the collected digital signals into image data.

5. A SPECT data processing method based on the NVIDIA AGX ORIN platform, characterized in that: The SPECT data processing system based on the NVIDIA AGX ORIN platform as described in any one of claims 1 to 4 is used to perform the following steps: Step S1: using the detector to collect analog data, and transmitting the collected analog data to the digital-to-analog conversion board; Step S2: The digital-to-analog conversion board converts the received analog data into digital signals, and transmits the digital signals to the data transfer board; Step S3: The data transfer board receives the digital signal and transmits the digital signal to the AGX data collection board; Step S4: the AGX data collection board receives the digital signal, processes the digital signal to generate image data, and transmits the generated image data to the host computer; Step S5: The host computer receives the image data and displays it.

6. The SPECT data processing method based on the NVIDIA AGX ORIN platform according to claim 5, characterized in that: The step S4 comprises: Step S4.1: preprocessing the received digital information to obtain preprocessed digital information; Step S4.2: Process the pre-processed digital information to obtain image data; Step S4.3: preprocessing the generated image data to obtain preprocessed image data; Step S4.4: Transmit the preprocessed image data to the host computer.

7. The SPECT data processing method based on the NVIDIA AGX ORIN platform according to claim 6, characterized in that: The step S4.1 includes: performing preprocessing on the received digital information including removing noise and abnormal values ​​to obtain preprocessed digital information.

8. The SPECT data processing method based on the NVIDIA AGX ORIN platform according to claim 6, characterized in that: The step S4.3 comprises: optimizing the generated image data according to the deep learning algorithm; The optimization of the generated image data according to the deep learning network includes: using the deep learning network to perform noise reduction, contrast enhancement, pixel value reconstruction of missing areas, image segmentation, lesion area identification and position marking on the generated images to obtain optimized image data.

9. The SPECT data processing method based on the NVIDIA AGX ORIN platform according to claim 5, characterized in that: The AGX data collection board realizes real-time streaming processing of image data based on the multi-task parallel processing capability of the NVIDIA AGX ORIN chip.

10. The SPECT data processing method based on the NVIDIA AGX ORIN platform according to claim 5, characterized in that: The data transfer board adds a frame header, version number, data length, data load, frame tail and checksum to the output data; when the AGX data collection board receives a digital signal, it completes the confirmation of the transmitted data including the frame header, version number, data length, data load and frame tail to obtain a checksum. When the currently obtained checksum is consistent with the checksum added by the data transfer board, the data verification is successful and the data is not lost or damaged.

Images