A method for determining brain PET-normalized reconstruction parameters based on clinical brain MRI and PET data

By using a virtual simulation process based on clinical brain MRI and PET data, standardized reconstruction parameters for brain PET were determined, solving the problems of complex operation and high cost in existing technologies. This enabled flexible and efficient image standardization, making it suitable for multi-center studies and clinical applications.

CN120807675BActive Publication Date: 2026-11-10CHINA JAPAN FRIENDSHIP HOSPITAL +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510820807.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-06-19
Publication Date
2026-11-10
Estimated Expiration
2045-06-19

AI Technical Summary

Technical Problem

The existing technology of using physical phantoms to determine standardized reconstruction parameters for brain PET is complex, costly, and limited in clinical application. It is also difficult to adapt to different data acquisition scenarios, thus restricting the standardized application of brain PET technology.

Method used

Based on clinical brain MRI and PET data, standardized reconstruction parameters suitable for specific PET equipment were determined through image processing and virtual simulation processes. These parameters included iterative algorithms, number of iterations, and filter types. Partial volume correction and activity mapping were performed using MRI segmentation results to optimize the activity distribution map. The reconstruction parameters were iteratively adjusted until they met the standardized requirements.

Benefits of technology

It simplifies the operation process, reduces costs, improves the clinical applicability and flexibility of reconstruction parameters, promotes multi-center studies and clinical standardization, and achieves image standardization across different devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120807675B_ABST
    Figure CN120807675B_ABST
Patent Text Reader

Abstract

This invention discloses a method for determining standardized reconstruction parameters for brain PET based on clinical brain MRI and PET data, relating to the field of medical image processing. First, paired clinical MRI and PET images are collected and preprocessed. Combining a partial volume correction algorithm and system parameters of the equipment, virtual reconstruction is performed to estimate the "real" brain activity distribution, resulting in a first simulated PET image. The difference or similarity between this simulated image and the clinical PET image is calculated. After multiple iterations, the optimal "real" brain activity distribution map is obtained and processed using a standardized Gaussian filter to obtain a "standardized PET image." Then, different candidate combinations of reconstruction parameters are constructed, and the optimal "real" brain activity distribution map is input. The second simulated PET image generated under each combination is iterated, and its difference or similarity is calculated with the "standardized PET image." The candidate reconstruction parameters with the smallest difference or highest similarity are selected as the final standardized reconstruction parameters. This invention eliminates the dependence on physical phantoms.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical image processing, specifically a method for determining standardized reconstruction parameters of brain PET based on clinical brain MRI and PET data. Background Technology

[0002] Brain PET imaging is an indispensable imaging technique for the diagnosis, staging, and efficacy evaluation of neurological diseases (such as Alzheimer's disease, Parkinson's disease, epilepsy, and brain tumors), as well as for brain function research. It reflects physiological and biochemical information such as brain tissue metabolism, blood flow, and neurotransmitter receptors by detecting the distribution of radioactive tracers in the brain.

[0003] However, different PET scanning devices (from different manufacturers or different models from the same manufacturer), different reconstruction algorithms, and parameter settings can lead to significant differences in the spatial resolution, noise level, and quantitative accuracy of the obtained PET images. These differences severely hinder direct comparison and data aggregation of PET images between different medical institutions and at different time points, limiting the conduct of large-scale multicenter clinical studies, clinical quantitative evaluation, and the development of standardized clinical guidelines.

[0004] like Figure 1 As shown: In multicenter PET imaging, even when faced with the same “true” radioactivity distribution (the brain image on the far left in the figure), the final PET images (the two brain images in the upper and lower middle of the figure) obtained after data acquisition and image reconstruction using PET systems at different sites (or devices) (here illustrated as “Site A PET acquisition and reconstruction” and “Site B PET acquisition and reconstruction”) will have significant differences in visual characteristics (such as sharpness and noise level).

[0005] The results differ even within the same site using different reconstruction parameters: the two curves (black and orange) for site A show that even on the same device, changing the reconstruction parameters (e.g., changing the subset number from 5 to 10) results in different characteristics of image resolution as a function of filtering. The results also differ between sites using the same nominal reconstruction parameters: comparing the orange curve (i10s10) for site A and the blue curve (i10s10) for site B, even with the same number of iterations and subsets, the image resolution characteristics obtained from the two sites still differ. To achieve the target resolution (pink area), they may require different post-processing filter FWHM values.

[0006] To achieve the synergistic or standardization of brain PET images—that is, to make PET images from different sources comparable—a key step is to determine a set of "standardized" reconstruction parameters suitable for specific equipment and scanning protocols. Traditionally, this process has relied on physical phantoms, such as the Hoffman 3D brain phantom. Operators need to precisely fill the phantom with radionuclides according to specific procedures to simulate the activity distribution of the human brain, and then perform scanning and image reconstruction. The optimal reconstruction parameters are adjusted and selected by analyzing the quality of the phantom images (such as resolution, uniformity, contrast recovery coefficient, etc.) and comparing them with target standards.

[0007] Although the Hoffman phantom method is the recognized reference method, it has many inconveniences in actual clinical practice:

[0008] • Complex operation: The steps of precise filling of the mold, avoiding air bubbles, and activity calibration are cumbersome and time-consuming.

[0009] • High cost: It requires the purchase and maintenance of specialized phantoms, and each use involves handling radioactive materials.

[0010] • Difficulty in clinical implementation: It is difficult to implement it frequently in busy clinical departments as a routine quality control or parameter optimization method.

[0011] • Limited representativeness: Phantoms are, after all, idealized models and differ from the complex anatomical structures and physiological variations of the brains of real patients.

[0012] Therefore, there is an urgent need for a more convenient, efficient method that can make full use of real clinical data to determine standardized reconstruction parameters for brain PET, so as to get rid of dependence on physical phantoms and promote the widespread standardized application of brain PET technology.

[0013] Meanwhile, the sources of paired imaging data acquired clinically are diverse, including integrated PET / MRI equipment, independent MRI scans performed after PET / CT scans, or PET and MRI scans performed at different time points. There is an urgent need for an ideal standardized parameter determination method that can flexibly adapt to different data acquisition scenarios. Summary of the Invention

[0014] To overcome the problems of complexity, high cost, and limited clinical application in determining standardized brain PET reconstruction parameters using physical phantoms (such as Hoffman phantoms) in existing technologies, this invention provides a method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data. Utilizing actual clinical PET scan data from patients and paired MRI data, a specific image processing and virtual simulation process is employed to determine standardized reconstruction parameters suitable for a specific PET device. This method is adaptable to paired PET and MRI data from different sources.

[0015] The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data comprises the following steps:

[0016] Step 1: Given the initial reconstruction parameters, collect the patient's clinical PET scan images and paired T1W-MRI images, and perform brain structure segmentation on the T1W-MRI images to obtain the MRI segmentation results.

[0017] The reconstruction parameters include the iterative algorithm type, number of iterations, number of subsets, filter type, cutoff frequency, etc.

[0018] Brain structure segmentation refers to the precise segmentation of major brain tissue regions such as gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF), or more detailed brain regions defined by brain atlases.

[0019] Step 2: Using a partial volume correction algorithm, the registered clinical PET scan images are partially corrected for volume effect using the MRI segmentation results.

[0020] Step 3: The activity values ​​of the PET scan images after partial volume correction are mapped onto a high-resolution MRI spatial grid according to the brain regions defined by MRI segmentation, as an initial estimate of the "true" brain activity distribution.

[0021] Step 4: Based on the current “real” brain activity distribution map, and combined with the initial reconstruction parameters and the system parameters of the PET device, perform a PET image simulation reconstruction process to generate the first simulated PET image.

[0022] The system parameters of a PET device include system geometry, detector response function / point spread function model, attenuation map, scattering correction model, etc.

[0023] Step 5: Compare the first simulated PET image with the real clinical PET scan image, calculate the difference or similarity between the two, and adjust the activity value in the "real" brain activity distribution map according to the difference using an optimization algorithm.

[0024] Algorithms for calculating differences or similarities include: mean squared error, normalized mutual information, structural similarity index, etc., as objective functions;

[0025] Optimization algorithms include: gradient descent, conjugate gradient, and algorithms within the expectation maximization (EM) framework;

[0026] The goal of the adjustment is to minimize the difference between the simulated PET images generated by the next virtual scan and the real clinical PET images.

[0027] Step 6: Return to Step 4 and repeat the simulation reconstruction. Perform multiple iterations until the difference between the simulated PET image and the real PET image is less than the preset convergence threshold, or the preset maximum number of iterations is reached, to obtain the optimal "real" brain activity distribution map.

[0028] Step 7: Convert the optimal "true" brain activity distribution map into a "normalized PET image" by applying Gaussian filtering (normalization processing - i.e. effective spatial resolution);

[0029] Standardization requirements typically refer to the requirement that the spatial resolution of an image should meet a recognized standard; for example, the resolution recommended by the European Association for Nuclear Medicine's EARL standard, or the uniform resolution defined in other multicenter research protocols.

[0030] Step 8: Construct different candidate combinations of reconstruction parameters, and use the optimal "real" brain activity distribution map as the input source image. Iterate through each group of candidate reconstruction parameters and combine them with the system parameters of the PET device to perform simulation reconstruction and generate the corresponding second simulated PET image for each group.

[0031] Step 9: Compare each set of corresponding second simulated PET images with the "standardized PET images", calculate the difference or similarity between the two, and select the set of candidate reconstruction parameters with the smallest difference or the highest similarity. These are the standardized reconstruction parameters that this PET device should use for this type of clinical brain scan.

[0032] The advantages of this invention are:

[0033] 1. A method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data, which eliminates the need for physical phantoms and completely eliminates the dependence on physical phantoms such as Hoffman phantoms. This significantly simplifies the operation process, reduces implementation costs and human resource requirements, and makes it possible to determine or validate standardized parameters in routine clinical workflows.

[0034] 2. A method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data. This method utilizes real clinical data, directly using the patient's actual scan data (including PET and paired MRI), making the final determined reconstruction parameters closer to actual clinical applications. This method may be more clinically applicable and representative than parameters obtained based on idealized phantoms.

[0035] 3. A method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data, which is highly flexible and can adapt to paired PET and MRI data from different sources, whether it is data from a PET / MRI integrated machine or combined data of PET / CT and independent MRI.

[0036] 4. A method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data has great potential for automation. The entire processing flow, from data input to parameter output, has a high potential for automation and can be implemented through software integration, thereby improving work efficiency and the objectivity of results.

[0037] 5. A method for determining standardized reconstruction parameters of brain PET based on clinical data to promote multicenter studies and clinical standardization: It provides a convenient and effective technical means for standardizing and coordinating brain PET images among different medical institutions and different PET devices, which helps to improve the data quality and credibility of multicenter clinical studies and promote the establishment of standardized clinical practice guidelines. Attached Figure Description

[0038] Figure 1 PET images reconstructed from data collected by PET systems at different sites in the prior art of this invention;

[0039] Figure 2 This is a schematic diagram of the method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data according to the present invention.

[0040] Figure 3 This is a flowchart of the method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data according to the present invention. Detailed Implementation

[0041] The embodiments of the present invention will now be described in complete and detailed manner with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] This invention provides a method for determining standardized reconstruction parameters of brain PET based on clinical brain MRI and PET data. It does not require a physical phantom and utilizes the patient's clinical brain positron emission tomography (PET) scan data and paired magnetic resonance imaging (MRI) data to determine standardized reconstruction parameters of PET images through a specific processing procedure. This method is suitable for promoting the collaboration and standardization of brain PET imaging in multi-center studies and clinical applications.

[0043] like Figure 2 As shown, the process mainly includes three steps: preprocessing of PET and paired MRI; virtual scanning and reconstruction of PET; and solving for PET reconstruction parameters that meet the standardization requirements. Specifically: First, using the patient's clinical PET scan data and paired high-resolution T1-weighted MRI (T1W-MRI) images, the MRI segmentation results are used as prior anatomical constraints for brain structure and activity distribution. Second, by combining a partial volume correction algorithm and iterative virtual reconstruction (or virtual scanning) based on device system parameters, an approximately "real," high-resolution brain radioactivity distribution is derived from the clinical PET data. Finally, based on this "real" activity distribution, parametric virtual scanning is performed again and compared with predefined standardized brain PET image targets. The reconstruction parameters are then iterated and optimized to determine a set of reconstruction parameters that best approximates the standardized target.

[0044] The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data, such as Figure 3 As shown, the specific steps are as follows:

[0045] Step 1: Given the initial reconstruction parameters, collect the patient's clinical PET scan images and paired T1W-MRI images, and perform brain structure segmentation on the T1W-MRI images to obtain the MRI segmentation results.

[0046] The reconstruction parameters include the iterative algorithm type, number of iterations, number of subsets, filter type, cutoff frequency, etc.

[0047] The specific steps are as follows:

[0048] 1.1 Acquiring the patient’s clinical brain PET imaging data includes images reconstructed using the standard acquisition protocol and clinically common (or default) reconstruction parameters of the current PET equipment (e.g., PET / CT or PET / MRI scanner).

[0049] 1.2. High-resolution T1-weighted MRI (T1W-MRI) images of the patient paired with the PET images were obtained from:

[0050] Acquired simultaneously with PET (e.g., on a PET / MRI device);

[0051] Unlike PET, MRI is performed at similar time points (e.g., after a PET / CT scan, the patient undergoes a separate MRI scan).

[0052] Archived MRI images acquired at different time points than PET, but which can be aligned with PET images using reliable image registration techniques.

[0053] 1.3. Accurately register the acquired clinical PET images to their paired T1W-MRI image space.

[0054] 1.4. Use the T1W-MRI images to segment brain structures: accurately segment major brain tissue regions such as gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF), or brain regions defined by more detailed brain atlases.

[0055] MRI segmentation can utilize tools or algorithms provided by mature brain imaging analysis software packages (such as FSL's FAST, SPM, FreeSurfer, etc.); registration between PET and MRI can employ rigid or affine registration algorithms based on measures such as mutual information and normalized mutual information; importantly, it is crucial to ensure the spatial consistency of PET and MRI data. These segmentation results will serve as anatomical prior constraints for activity distribution estimation and partial volume effect correction in subsequent steps.

[0056] Step 2: Using the Partial Volume Correction (PVC) algorithm, the registered clinical PET scan images are partially corrected for volume effect using the MRI segmentation results.

[0057] The application of partial volume correction is one of the key steps. Its purpose is to use the high-resolution anatomical information provided by MRI to improve the quantitative accuracy of PET images, provide a more reliable initial activity estimate for subsequent iterations, and serve as an anatomical constraint.

[0058] Using MRI segmentation results (i.e. anatomical constraints), partial volume effect correction was performed on the registered clinical PET images. This correction aims to compensate for the diffusion and mixing effects of signals between different tissues caused by the limited spatial resolution of the PET system, thereby more accurately estimating the true mean activity in each brain region.

[0059] Algorithms that can be used for partial volume correction in brain PET scans include, but are not limited to:

[0060] The method based on the geometric transition matrix (GTM) proposed by Meltzer et al. (Meltzer CC, et al. J NuclMed. 1990;21(4):679-88);

[0061] The two-parameter (MG) or three-parameter method proposed by Muller-Gartner et al. (Muller-Gartner HW, et al. J Cereb Blood Flow Metab. 1992;12(4):571-83);

[0062] Rousset et al. proposed a region-based iterative deconvolution method (Rousset OG, et al. J NuclMed. 1998;39(5):904-11);

[0063] Van Cittert's iterative deconvolution method and its improved version for PET;

[0064] Correction methods based on accurate modeling of the point spread function (PSF), such as variants of the Richardson-Lucy deconvolution algorithm;

[0065] A machine learning or deep learning-driven approach to PVC using MRI priors.

[0066] Step 3: The activity values ​​of the PET scan images after partial volume correction are mapped onto a high-resolution MRI spatial grid according to the brain regions defined by MRI segmentation, as an initial estimate of the "true" brain activity distribution.

[0067] By combining high-resolution MRI anatomical information with clinical PET images that are affected by partial volume effects and have limited resolution, a high-resolution brain radioactivity distribution map that is as close as possible to physiological reality can be derived.

[0068] Step 4: Based on the current “real” brain activity distribution map, and combined with the actual reconstruction parameters used when the original PET images were acquired and the system parameters of the PET device, a PET image simulation reconstruction process is performed to generate the first simulated PET image.

[0069] The system parameters of a PET device include the detector response function / point spread function (PSF), system geometry, attenuation map generation method or actual attenuation data, scattering correction model and parameters, etc.

[0070] The virtual scanning (or simulation reconstruction) process is a PET image simulation reconstruction process based on the approximate "real" brain activity distribution, the reconstruction parameters to be evaluated, and the system parameter model of the PET device; it is also a simulation of the physical imaging process of the PET scanner. Its core is the forward projection operation, which projects a given activity distribution map according to the system model (including geometric, blurring, attenuation, scattering, and other effects) to obtain simulated detector data (sinogram), and then obtains a simulated image through a reconstruction algorithm (usually consistent with or similar to the reconstruction algorithm actually used by the device); an accurate device system matrix or point spread function model is the key to ensuring simulation accuracy.

[0071] Step 5: Compare the first simulated PET image with the real clinical PET scan image, calculate the difference or similarity between the two, and adjust the activity value in the "real" brain activity distribution map according to the difference using an optimization algorithm.

[0072] Algorithms for calculating differences or similarities include: mean squared error, normalized mean squared error, mutual information, structural similarity index, etc., as objective functions;

[0073] Optimization algorithms include: gradient descent, conjugate gradient, and algorithms within the expectation maximization (EM) framework;

[0074] The goal of the adjustment is to minimize the difference between the simulated PET images generated by the next virtual scan and the real clinical PET images.

[0075] Step 6: Return to Step 4 and repeat the simulation reconstruction. Perform multiple iterations until the difference between the simulated PET image and the real PET image is less than the preset convergence threshold, or the preset maximum number of iterations is reached. The optimal high-resolution activity distribution map is considered to be an approximate "real" brain activity distribution map.

[0076] The choice of iterative optimization algorithm can be varied, such as gradient-based optimization methods or iterative update strategies based on the expectation maximization (EM) concept, with the goal of minimizing the difference between the simulation reconstruction results and the actual clinical scan results.

[0077] Step 7: Convert the optimal "true" brain activity distribution map into a "normalized PET image" by applying Gaussian filtering (normalization processing - i.e. effective spatial resolution);

[0078] Standardization requirements typically refer to the requirement that the spatial resolution of an image should meet a recognized standard, such as the resolution recommended by the European Association for Nuclear Medicine's EARL standard, or a uniform resolution defined in other multicenter research protocols.

[0079] Standardized targets (such as target spatial resolution FWHM values) need to be defined in advance based on industry standards, multicenter study protocols, or specific clinical needs.

[0080] Gaussian smoothing is a common and effective method for degrading high-resolution, near-"true" activity maps to images that conform to specific lower-resolution standards.

[0081] Step 8: Construct different candidate combinations of reconstruction parameters, and use the optimal "real" brain activity distribution map as the input source image. Iterate through each group of candidate reconstruction parameters and combine them with the system parameters of the PET device to perform simulation reconstruction and generate the corresponding second simulated PET image for each group.

[0082] The traversal range and step size of the reconstruction parameters should be reasonably set based on equipment characteristics, clinical experience, and computational resources, in order to find the optimal or near-optimal parameter combination within an acceptable computation time. A search space for PET reconstruction parameters to be optimized should be defined, including different reconstruction algorithms (such as OSEM, FBP), different combinations of iteration numbers and subsets, and different post-processing filter types and their parameters (such as the FWHM Gaussian filter). For each candidate reconstruction parameter combination in this parameter space, an approximately "real" brain activity distribution is used as the input source image. Combining the current candidate reconstruction parameters with the system parameters of the PET device, a virtual scan (simulation reconstruction) is performed again to generate a new simulated PET image (referred to as the second simulated PET image).

[0083] Step 9: Compare the second simulated PET image obtained by each set of candidate reconstruction parameters with the "standardized PET image", calculate the difference or similarity between the two, and select the set of candidate reconstruction parameters with the smallest difference or the highest similarity. This set of candidate reconstruction parameters is the standardized reconstruction parameter that this PET device should use for this type of clinical brain scan.

[0084] The metric for minimizing differences can be the norm of the differences in image pixel values ​​(such as the L2 norm, i.e., the root mean square error RMSE), image structural similarity indices (such as SSIM), or differences in quantitative indices within a specific region of interest.

[0085] This invention begins with acquiring clinical PET and paired MRI data from patients, followed by data preprocessing, partial volume correction using MRI priors, and solving for an approximate "real" brain activity distribution through iterative virtual scanning. Finally, based on this "real" activity distribution, standardized PET reconstruction parameters are determined through further parameterized virtual scanning. This invention addresses the technical problem of how to determine standardized brain PET (positron emission tomography) reconstruction parameters using clinical data without the need for physical phantoms.

[0086] Example:

[0087] a) Clinical data acquisition, preprocessing, and extraction of anatomical constraints:

[0088] Acquire the patient's clinical brain PET image data, acquire high-resolution MRI (magnetic resonance imaging) image data paired with the PET images, register the PET images to the space of the MRI images, and segment different brain regions based on the MRI images as anatomical constraints.

[0089] b) Iteratively solve for an approximate "real" brain activity distribution:

[0090] Using the brain region segmentation results obtained in step a) as a constraint, partial volume correction is performed on the registered PET image. Based on the activity distribution after partial volume correction, an initial estimate is made. Combined with the reconstruction parameters and PET equipment system parameters corresponding to the original PET acquisition, a first virtual scan is performed to generate a first simulated PET image. The "real" activity distribution is iteratively adjusted through an optimization algorithm to minimize the difference between the first simulated PET image and the clinical brain PET image, thereby obtaining an approximate "real" brain activity distribution.

[0091] c) Steps for determining standardized reconstruction parameters:

[0092] Based on the approximate "real" brain activity distribution obtained in step b), a standardized target PET image that meets the predefined standardization requirements is generated. Then, multiple sets of candidate reconstruction parameters within a preset range are traversed. For each set of candidate reconstruction parameters, the approximate "real" brain activity distribution is used in combination with the current candidate reconstruction parameters and the PET device system parameters to perform a second virtual scan to generate a second simulated PET image. The candidate reconstruction parameter combination that minimizes the difference between the second simulated PET image and the standardized target PET image is selected as the standardized reconstruction parameter of the PET device.

[0093] The system for determining standardized brain PET reconstruction parameters based on clinical data, as described in this invention, includes the following processing modules:

[0094] The clinical data acquisition, preprocessing, and anatomical constraint extraction module is used to perform the operation in step a);

[0095] An iterative solution module for approximating the "real" brain activity distribution is used to perform the operation in step b);

[0096] The standardized reconstruction parameter determination module is used to perform the operation in step c).

[0097] The specific implementation method of "virtual scanning" or "simulation reconstruction" (such as Monte Carlo simulation or analytical projection / back projection and system response model) is not the core of this invention, as long as it can effectively simulate the PET imaging process.

Claims

1. A method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data, characterized in that, The specific steps are as follows: Step 1: Given the initial reconstruction parameters, collect the patient's clinical PET scan images and paired T1W-MRI images, and perform brain structure segmentation on the T1W-MRI images to obtain the MRI segmentation results; Step 2: Using a partial volume correction algorithm, the registered clinical PET scan images are partially corrected for volume effect using the MRI segmentation results; Step 3: The activity values ​​of the PET scan images after partial volume correction are mapped onto a high-resolution MRI spatial grid according to the brain regions defined by MRI segmentation, as an initial estimate of the "true" brain activity distribution. Step 4: Based on the current "real" brain activity distribution map, and combined with the initial reconstruction parameters and the system parameters of the PET device, perform a simulation reconstruction process of the PET image to generate the first simulated PET image. Step 5: Compare the first simulated PET image with the real clinical PET scan image, calculate the difference or similarity between the two, and adjust the activity value in the "real" brain activity distribution map according to the difference using an optimization algorithm. Step 6: Return to Step 4 and repeat the simulation reconstruction. Perform multiple iterations until the difference between the simulated PET image and the real PET image is less than the preset convergence threshold, or the preset maximum number of iterations is reached, to obtain the optimal "real" brain activity distribution map. Step 7: Convert the optimal "real" brain activity distribution map into a "normalized PET image" by applying a Gaussian filter. Standardization requirements refer to the requirement that the spatial resolution of the image should meet a standard; this standard includes the resolution recommended by the European Association for Nuclear Medicine's EARL standard, or a uniform resolution defined in other multicenter study protocols. Step 8: Construct different candidate combinations of reconstruction parameters, and use the optimal "real" brain activity distribution map as the input source image. Iterate through each group of candidate reconstruction parameters and combine them with the system parameters of the PET device to perform simulation reconstruction and generate the corresponding second simulated PET image for each group. Step 9: Compare each set of corresponding second simulated PET images with the "standardized PET images", calculate the difference or similarity between the two, and select the set of candidate reconstruction parameters with the smallest difference or the highest similarity. These are the standardized reconstruction parameters that this PET device should use for clinical brain scans.

2. The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data as described in claim 1, characterized in that, The reconstruction parameters include iterative algorithm type, number of iterations, number of subsets, filter type, and cutoff frequency; Brain structure segmentation refers to the precise segmentation of gray matter (GM), white matter (WM), brain tissue regions, or brain regions defined by brain atlases.

3. The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data as described in claim 1, characterized in that, The system parameters of the PET equipment include system geometry, detector response function / point spread function model, attenuation map, and scattering correction model.

4. The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data as described in claim 1, characterized in that, The algorithm for calculating differences or similarities includes using mean squared error, normalized mutual information, or structural similarity index as the objective function.

5. The method for determining standardized brain PET reconstruction parameters based on clinical brain MRI and PET data as described in claim 1, characterized in that, The goal of the adjustment is to minimize the difference between the simulated PET images generated by the next virtual scan and the real clinical PET images.

Citation Information

Patent Citations

  • PET image reconstruction method and system, readable storage medium and equipment

    CN109712213A

  • PET reconstruction method based on anatomical prior information

    CN111402357A