Self-adaptive PET real-time imaging method based on fine system matrix

Abstract

The invention discloses a self-adaptive PET (positron emission tomography) real-time imaging method based on a fine system matrix. The self-adaptive PET real-time imaging method comprises the following steps: 1) simulating and generating the fine system matrix of a PET system; 2) based on coincidence data acquired by the PET system and the fine system matrix, performing image reconstruction by adopting an ordered subset maximum expected value method; wherein in the iteration process of image reconstruction, subset division and iteration updating are carried out on the coincidence data, the step length of each iteration updating is controlled according to a relaxation factor alpha (t) which adaptively changes along with time or a relaxation factor alpha (n) which adaptively changes along with activity, and image reconstruction is carried out according to each subset after each iteration updating. According to the invention, the distribution of the radiopharmaceutical in the scanned object can be observed in real time, a PET real-time imaging image is directly obtained, and a 4D imaging image is generated after scanning is finished.

Description

Technical Field

[0001] The invention belongs to the field of nuclear medicine imaging and high-performance parallel computing, and relates to a PET real-time imaging method, in particular to a fast adaptive PET real-time imaging method based on a fine system matrix. Background Art

[0002] With the development of nuclear medicine imaging technology, positron emission tomography (PET) as a functional molecular imaging technology provides advanced imaging means for scientific research, clinical diagnosis and treatment. Tracers labeled with positron decay radioisotopes participate in metabolism. By detecting the gamma photons generated by positron annihilation, the distribution of radionuclides in the tissues and organs of the scanned object can be imaged, and the metabolic process of radioactive drugs can be imaged. This provides data basis for drug metabolism mechanisms and tissue metabolism levels. Unlike structural imaging methods, PET imaging has the ability to detect early small lesions and non-organic lesions, which is of great significance for the early detection and treatment of major diseases such as tumors. Therefore, PET imaging plays an important role in preclinical research, clinical diagnosis and treatment.

[0003] In the PET imaging system, there are detector subsystems, electronics subsystems, data processing subsystems and image processing subsystems. The detector subsystem performs energy deposition detection on photons with a certain energy, and the electronics subsystem obtains effective signals, and then outputs the event signal containing the deposition energy, deposition position and deposition time to the data processing subsystem. The data processing subsystem organizes the detection events into Listmode data or Sinogram data in a specific format. The image processing subsystem performs data correction and image reconstruction based on Listmode data or Sinogram data to obtain a radionuclide distribution image. In the current PET system design, positron annihilation is usually considered as the main way to produce photon pairs. Therefore, an effective PET detection event contains the information of a pair of photons, which is also called a coincidence response line (LOR). A large number of detected coincidence response lines constitute Listmode data. Listmode data contains information such as the energy, time, and position of the annihilation photons, while Sinogram data organizes the detected coincidence response lines into a fixed-size data matrix through the positional relationship of LOR. Therefore, in PET data processing, Listmode data varies with the activity of the radiation source and the acquisition time, and Sinogram data is an array of fixed size. With the research and development of PET data processing algorithms, reconstruction methods for Listmode data format or Sinogram data format have been applied. In terms of imaging algorithms, it mainly includes analytical algorithm imaging and iterative algorithm imaging.

[0004] PET imaging methods are usually divided into static imaging and dynamic imaging. Static imaging generally refers to data collection and data reconstruction after the distribution of radioactive drugs in the scanned object is close to stable. This imaging is 3D imaging. Dynamic imaging refers to the acquisition and data processing mode that takes into account the distribution changes of radioactive drugs over time. Data collection starts before the scanned object is injected with drugs. After data collection is completed, the data is divided into time periods according to metabolic characteristics for 4D imaging, that is, dynamic imaging.

[0005] Traditionally, whether it is static imaging or dynamic imaging, data is collected first and then processed offline. However, in actual scientific research or clinical applications, it is necessary to obtain the distribution of radioactive drugs online in real time. The high-precision real-time imaging of the PET system is of great significance for the real-time adjustment of the scanning plan, the real-time control of the dose during radiotherapy, and the improvement of the scanning flux.

[0006] PET system imaging is based on Listmode data or Sinogram data processing, and the response line is imaged through analytical or iterative methods to obtain the distribution image of the radioactive drug. The existing method is performed offline, and data processing is required after data acquisition is completed. It is impossible to obtain the distribution image of the radioactive drug in real time, or only a relatively fuzzy imaging image can be given, which cannot meet the scene requirements of real-time imaging, such as the preliminary judgment of the drug metabolism model and the dose assessment during radiotherapy. Summary of the invention

[0007] In view of the problems existing in the prior art, the purpose of the present invention is to provide a fast adaptive PET real-time imaging method based on a fine system matrix, so that during the PET scanning process, the distribution of radioactive drugs in the scanned object can be observed in real time, PET real-time imaging images can be directly obtained, and 4D imaging images can be generated after the scanning is completed.

[0008] The present invention provides real-time dynamic data processing and adaptive parameter adjustment according to the scanning target during the scanning process of the PET device. Real-time update of millisecond-level imaging during the scanning process is achieved. On the basis of providing real-time imaging of the current radioactive drug distribution, 4D dynamic imaging is obtained after the scanning is completed. The imaging time is greatly reduced, and the imaging efficiency and real-time performance are improved. This technology brings a new real-time reconstruction strategy to the field of PET imaging, thereby promoting the real-time and efficient application of PET imaging.

[0009] The technical solution of the present invention is:

[0010] An adaptive PET real-time imaging method based on a fine system matrix comprises the following steps:

[0011] 1) Simulate and generate the detailed system matrix of the PET system;

[0012] 2) Based on the coincident data collected by the PET system and the fine system matrix, an ordered subset maximum expectation method is used to perform image reconstruction; wherein in the iterative process of image reconstruction, the coincident data is divided into subsets and iteratively updated, and the update step size of each iteration is controlled according to a relaxation factor α(t) that adaptively changes with time or a relaxation factor α(n) that adaptively changes with activity, and image reconstruction is performed separately according to each subset after each iterative update.

[0013] Furthermore, the ordered subset maximum expected value method uses the formula Determine the distribution of the drug in the body in the image reconstructed using the jth subset at the k+1th iteration The system matrix of the PET system is obtained by simulation, and the element ρ in the system matrix is ij After ray driving, the corresponding element a in the fine system matrix is ​​obtained i,j , ρ ij is the probability of the jth LOR of the data being detected by the ith simulated voxel radiation source, I is the total number of pixels in the field of view, and J is the total number of LORs of the data being detected by the ith simulated voxel radiation source.

[0014] Furthermore, the ordered subset maximum expected value method uses the formula Determine the distribution of the drug in the body in the image reconstructed using the jth subset at the k+1th iteration The system matrix of the PET system is obtained by simulation, and the element ρ in the system matrix is ij After ray driving, the corresponding element a in the fine system matrix is ​​obtained i,j , ρ ij is the probability of the jth LOR of the data being detected by the ith simulated voxel radiation source, I is the total number of pixels in the field of view, and J is the total number of LORs of the data being detected by the ith simulated voxel radiation source.

[0015] Furthermore, a GPU parallel acceleration solution or a CPU multi-core parallel method is used to subset the conforming data and iteratively update the subsets.

[0016] Furthermore, a histogram equalization method is used to process each reconstructed image to improve the contrast of the reconstructed image.

[0017] Furthermore, a Monte Carlo simulation method is used to generate a detailed system matrix of the PET system.

[0018] Furthermore, the fine system matrix is ​​a point drive system matrix or a ray drive system matrix; the storage format of the conforming data is Listmode data or Sinogram data.

[0019] The advantages of the present invention are as follows:

[0020] (1) PET is used for radiotherapy dose monitoring, providing real-time drug distribution;

[0021] (2) Adaptively change the subset update mechanism based on the application scenario to better meet the imaging objectives;

[0022] (3) Realize online 4D dynamic reconstruction and improve the utilization rate of PET equipment;

[0023] (4) The method is applicable to all equipment and does not depend on specific models. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is the overall flow chart of the present invention.

[0025] Figure 2 This is the matrix distribution diagram of the single simulation point system.

[0026] Figure 3 Generate a flow chart for the system matrix.

[0027] Figure 4 Comparison before and after histogram equalization processing;

[0028] (a) Before treatment, (b) After treatment.

[0029] Figure 5 The flowchart of real-time imaging based on time framing.

[0030] Figure 6 The flowchart of real-time imaging based on counting and framing. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention but not to limit the scope of the present invention.

[0032] The present invention designs a fast adaptive PET real-time imaging method based on a fine system matrix, which realizes simultaneous acquisition and imaging, and image reconstruction during data acquisition, thereby achieving real-time monitoring and evaluation of the distribution of radioactive drugs in the scanned object. The overall process of the method is as follows Figure 1 As shown, it mainly includes:

[0033] 1) Generation of detailed system matrix based on Sinogram data storage format or Listmode data storage format;

[0034] 2) Adaptive subset update mechanism;

[0035] 3) GPU-based parallel acceleration processing;

[0036] 4) Adaptive image display based on image statistical histogram.

[0037] 1. Fine system matrix generation method

[0038] In PET imaging, the system matrix characterizes the physical relationship between the system radiation source distribution and the projection data. The system matrix describes the probability of detecting the LOR formed by the uniform emission of Gamma rays into the space by the pixels filled with radioactive drugs in the imaging space. There are analytical methods, simulation methods and experimental methods to generate the system matrix. In the present invention, the system matrix is ​​generated by Monte Carlo simulation, which has higher accuracy than the analytical method and higher generation efficiency than the experimental method. The simulation uses the axial translation symmetry and axial rotation symmetry of the PET detection system to generate a refined system matrix. Set the simulation nuclear area, simulate and store the effective LOR event distribution point by point, such as Figure 2 As shown. This system matrix generation process uses pixels as the organizational structure, denoted as P-matrix, and can be used for pixel-driven image reconstruction. In ray-driven imaging applications, the response line LOR is used as the organizational structure, and the system matrix P-matrix obtained by simulation is used to drive the ray, and the ray-driven system matrix R-matrix is ​​generated as the refined system matrix. The process is as follows Figure 3 As shown. The probability of the jth LOR detected by the ith voxel radiation source in the simulated system matrix is ​​ρ ij , its expression is shown in 1. Where N i represents the simulated radiation source activity at the ith voxel, N ij It represents the detection count of the jth LOR detected by the i-th voxel radiation source.

[0039]

[0040] Then the element ρ in the system matrix ij The element value α after ray driving ij The expression of is shown in formula 2, where represents the total count of LOR j in the full field of view i∈(0,I), where I is the total number of pixels in the field of view of the PET system.

[0041]

[0042] The above system matrix is ​​divided into point-driven system matrix matrix and ray-driven system matrix R-matrix. Based on the characteristics of data storage, Sinogram has pixel attributes and Listmode has ray attributes. Generally, matrix is ​​used in Sinogram data organization and reconstruction process, and R-matrix is ​​used in Listmode data organization and reconstruction process.

[0043] 2. Adaptive subset update mechanism (subset partitioning and relaxation factor, two modes)

[0044] The PET system collects coincident data in real time and reconstructs the coincident data through the reconstruction algorithm. In order to meet the requirements of high-precision real-time imaging, this method uses the ordered subsets expectation maximum (OSEM) method for image reconstruction. In the iterative process, the coincident data is divided into subsets for iterative updates. In the update process, adaptive subset setting and update strategies are performed according to the count rate estimation and application objectives. Two update strategies are proposed, the time-based T-mode real-time imaging mode and the count-based N-mode real-time imaging mode. This method allows the algorithm to first capture the rapid changes in drug distribution and then gradually transition to capturing slower changes, which helps to more accurately reconstruct the dynamic process of drugs in the body. The relaxation coefficient is a parameter used in the iterative algorithm to control the update step size of each iteration. This study adopts two update methods, introducing a relaxation factor α(t) that changes adaptively with time and a relaxation factor α(n) that changes adaptively with activity. A lookup table of the influence factor λ is formulated according to the drug type and the metabolic sensitivity of the organ of interest to the drug. The expression of the influence factor λ is formula (3).

[0045] α(t)=e -λt / T

[0046] α(n)=e -λn / N

[0047] λ=F(A,R)......(3)

[0048] Where A is the type of radioactive drug, and R is the estimated drug uptake in the region of interest. For drug one-time injection metabolic imaging, which has a greater impact on time changes, the relaxation factor of T-mode changes with the metabolism of the radioactive drug. When the drug distribution changes rapidly, α(t) will be larger, which will speed up the iterative convergence. As time goes by, the change in drug distribution slows down, and α(t) gradually decreases to ensure the stability of the algorithm.

[0049] In the metabolic stable interval, the subsets can be divided according to the number of cases or acquisition time with the uniformity over time. In one iteration, the data of each subset will update the image once. Then, under the same amount of data and the same number of iterations, the method of dividing the subsets will update the image more times, increase the image convergence speed, and make the image closer to the real image distribution.

[0050] 3. GPU-based parallel acceleration

[0051] This method uses GPU parallel acceleration solutions for timing subset or quantitative subset processing and CUDA multi-threaded parallel computing to increase the speed of image reconstruction and meet clinical application requirements.

[0052] 4. Adaptive contrast image display based on statistical histogram

[0053] After the image is reconstructed in real time, this study uses a method of timed image acquisition to refresh the reconstructed image in the software interface at regular intervals, allowing users to monitor the dynamic distribution of drugs in the scanned object in real time. Due to the uneven physical characteristics of the tracer's distribution in the body, the radioactivity in different areas varies greatly, resulting in a wide range between the maximum and minimum values ​​in the image. In order to obtain better visual effects when displaying the image, the contrast of the image needs to be adjusted. This study uses a histogram equalization method to improve the contrast of PET images, thereby improving the diagnostic value of the image, such as Figure 4 shown.

[0054] 5. Update the image

[0055] In order to view the dynamic process of the distribution of radioactive drugs in the body, this study adopted a method of dynamic framing based on time, with the frame interval from small to large. This method allows the algorithm to first capture the rapid changes in drug distribution, and then gradually transition to capturing slower changes, which helps to more accurately reconstruct the dynamic process of drugs in the body. The relaxation coefficient (relaxation parameter) is a parameter used in the iterative algorithm to control the update step size of each iteration. This study adopted two update methods, introducing a relaxation factor α(t) that changes adaptively with time and α(n) that changes adaptively with activity. This means that in the early stages of the acquisition and iteration process, when the drug distribution changes faster, α(t) will be larger, which will speed up the iterative convergence. As time goes by, the changes in drug distribution slow down, and α(t) gradually decreases to ensure the stability of the algorithm.

[0056]

[0057] In this method, the system matrix can be obtained by simulation, analysis or experiment. The system matrix acquisition only affects the accuracy of this method. The storage format of the collected conforming data can use Listmode data or can be based on Sinogram data.

[0058] Example:

[0059] The real-time reconstruction test based on Listmode data proposed in this invention was carried out on the animal panoramic PET independently developed by the Institute of High Energy Physics of the Chinese Academy of Sciences. Detector related parameters: crystal bar (1.58×1.58×10mm 3 ), detector effective field of view (143×143×213mm3 ).

[0060] PET scans were performed on mice. While collecting data, reconstruction was performed based on the reconstruction method mentioned in this study. The software interface updated the current reconstructed image in real time, reflecting the real-time distribution of the current drug in the mouse. The results show that the real-time reconstruction method of the present invention can view the distribution of drugs in mice in real time.

[0061] The reconstruction algorithm in this scheme can also be an extended ordered subset expectation maximum method or other reconstruction algorithms; in the adaptive subset update mechanism, specific nuclides can generate different parameters due to different markers; data acceleration is not limited to GPU methods or CPU multi-core parallel methods.

[0062] Although the specific embodiments of the present invention are disclosed for the purpose of illustration, the purpose is to help understand the content of the present invention and implement it accordingly, those skilled in the art will understand that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the best embodiment, and the scope of the present invention is subject to the scope defined in the claims.

Claims

1. An adaptive PET real-time imaging method based on a fine system matrix, the steps comprising: 1) Simulate and generate the detailed system matrix of the PET system; 2) reconstructing an image using an ordered subset maximum expectation method based on the coincident data collected by the PET system and the fine system matrix; In the iterative process of image reconstruction, the conforming data is divided into subsets and iteratively updated, and the update step size of each iteration is controlled according to a relaxation factor α(t) that adaptively changes with time or a relaxation factor α(n) that adaptively changes with activity, and image reconstruction is performed on each subset after each iterative update.

2. The method according to claim 1, characterized in that: The ordered subset maximum expected value method uses the formula Determine the distribution of the drug in the body in the image reconstructed using the jth subset at the k+1th iteration The system matrix of the PET system is obtained by simulation, and the element ρ in the system matrix is ij After ray driving, the corresponding element a in the fine system matrix is ​​obtained i,j , ρ ij is the probability of the jth coincident data LOR detected by the ith simulated voxel radiation source, I is the total number of pixels in the field of view of the PET system, and J is the total number of coincident data LORs detected by the ith simulated voxel radiation source.

3. The method according to claim 1, characterized in that The ordered subset maximum expected value method uses the formula Determine the distribution of the drug in the body in the image reconstructed using the jth subset at the k+1th iteration The system matrix of the PET system is obtained by simulation, and the element ρ in the system matrix is ij After ray driving, the corresponding element a in the fine system matrix is ​​obtained i,j , ρ ij is the probability of the jth coincident data LOR detected by the ith simulated voxel radiation source, I is the total number of pixels in the field of view of the PET system, and J is the total number of coincident data LORs detected by the ith simulated voxel radiation source.

4. The method according to claim 1, 2 or 3, characterized in that: A GPU parallel acceleration solution or a CPU multi-core parallel method is used to subset the conforming data and iteratively update the subsets.

5. The method according to claim 1, 2 or 3, characterized in that: The histogram equalization method is used to process each reconstructed image to improve the contrast of the reconstructed image.

6. The method according to claim 1, 2 or 3, characterized in that: The Monte Carlo simulation method is used to generate the detailed system matrix of the PET system.

7. The method according to claim 1, 2 or 3, characterized in that: The detailed system matrix is ​​a point drive system matrix or a ray drive system matrix; the storage format of the conforming data is Listmode data or Sinogram data.

Images