A method for obtaining normalized correction factors for PET image reconstruction
Through the Poisson distribution principle and iterative solution method, the artifact problem caused by the uneven efficiency of the detection unit in PET image reconstruction is solved, and high-quality normalization correction and image reconstruction are achieved, which simplifies the operation process and reduces costs.
Patent Information
- Application Number
- CN202210358814.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-06
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-04-06
AI Technical Summary
In traditional PET image reconstruction, artifacts are caused by the uneven efficiency of the detection unit, and the traditional normalization correction method is complex, dangerous and costly, and error transmission affects image quality.
The objective function is constructed using the Poisson distribution principle, and the normalized correction factor is obtained through iterative solution. Combined with the MAP iterative reconstruction algorithm and gradient descent method, normalized correction of non-uniform radio source data is realized, and the image reconstruction process is fused to reduce error transmission. .
Improve image quality, simplify operational processes, reduce operational costs, and is suitable for a variety of PET detection systems without the need for additional uniform radiation sources and complex mechanical positioning equipment.
Smart Images

Figure CN114677455B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a method for obtaining a normalized correction factor for PET image reconstruction and a PET image reconstruction method. Background Art
[0002] Positron emission tomography (PET) is a high-end nuclear medicine imaging diagnostic device. In actual operation, radionuclides (such as 18 F. 11 C, etc.) to label the metabolites and inject the nuclides into the human body, and then use PET to perform functional metabolic imaging on the patient to reflect the status of life metabolic activities, thereby achieving the purpose of diagnosis.
[0003] The PET imaging system is usually a ring-shaped detection system containing tens of thousands of detection units. Affected by the geometric position and performance differences, such as crystal luminescence efficiency, crystal packaging, crystal coupling with PMT (photomultiplier tube) or silicon photomultiplier tube (SiPM), electronic system, different incident angles of photon pairs, etc., the detection efficiency of the detection unit is not consistent, resulting in its output not accurately reflecting the input photon beam intensity, which will inevitably introduce artifacts in the reconstruction process. In order to accurately model the detection system, the user must calibrate the detection efficiency of the detector in advance, that is, normalization correction.
[0004] Traditional normalization correction uses a uniform source placed at the center of the system (such as a rod source rotating at a constant speed, a cylindrical barrel source filled with FDG solution, or a solid cylindrical Ge barrel source) to uniformly illuminate each detection unit, and uses a model-based normalization method to process the detection data. The intrinsic detection efficiency of the crystal, the normalization factor affected by system geometric factors, and the normalization factor related to the counting rate are extracted from the data, and finally they are integrated into a normalization correction factor that reflects the overall detection efficiency, and the measurement data is normalized and corrected.
[0005] The detection efficiency of the detector will change slowly with the change of working conditions such as time, temperature, humidity, etc., and the replacement of the detector module will also cause a large change in the detection efficiency of the corresponding position. Therefore, normalization correction needs to be performed frequently to accurately reflect the changes in the performance of the PET system. However, the traditional normalization correction method will encounter the following problems in practical applications: the uniform source needs to be placed in the center of the detector to ensure that all detector units can be uniformly irradiated, which requires high position accuracy, and the operation is relatively complicated when fine-tuning the position. At the same time, the operator will also be at risk of radioactive exposure; the FDG barrel source needs to be manually filled, and this process also has the risk of radioactive exposure; whether using a solid Ge source or a FDG liquid source, it needs to be purchased additionally, and because the radioactive source will decay, it needs to be purchased repeatedly for normalization correction, which increases the operating cost; the axial field of view of the current PET system is getting larger and larger. In order to ensure that the radioactive source can still uniformly irradiate all detector crystals in the normalization test, the axial size of the radioactive source is designed to be longer and longer, which greatly increases the weight. At the same time, the positioning and moving operation becomes very difficult, which brings increasing challenges to the production and mechanical positioning of the uniform source; the traditional normalization correction algorithm usually uses variance reduction (Variance Reduction) Reduction) technology performs average noise reduction on the collected data and extracts the normalized correction factor from it. The normalized correction process is independent of the reconstruction process. Therefore, the errors caused by the modeling accuracy deviation and noise in the normalized correction process will be transferred to the image reconstruction, reducing the image quality and even causing image artifacts in severe cases. Summary of the invention
[0006] 1. Technical issues to be resolved
[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for obtaining a normalized correction factor for PET image reconstruction and a PET image reconstruction method.
[0008] (II) Technical solution
[0009] In order to achieve the above object, the main technical solutions adopted by the present invention include:
[0010] In a first aspect, an embodiment of the present invention provides a method for obtaining a normalized correction factor for PET image reconstruction, comprising:
[0011] S10, acquiring PET detection data y for PET image reconstruction and a predetermined normalized correction factor for PET image reconstruction and an initial value of radioactivity distribution x; the PET detection data is not required to be a uniform radiation central source, and data collected by a detection device of a non-uniform radiation source or a non-centrally positioned radiation source may be selected;
[0012] S20, constructing an objective function with unknowns being a normalized correction factor and a radioactivity distribution x according to the PET detection data and based on the principle of Poisson distribution, wherein the objective function is composed of a log-likelihood function with the normalized correction factor and the radioactivity distribution x as unknowns and a penalty function R(x) with the radioactivity distribution x as an unknown;
[0013] S30, iteratively solving the objective function according to the normalized correction factor and the initial value of the radioactivity distribution x, so that the normalized correction factor converges to the maximum likelihood estimation value, and obtaining the iterative normalized correction factor as the normalized correction factor for PET image reconstruction;
[0014] Wherein, the maximum likelihood estimate is a value obtained by performing maximum likelihood estimation on the log-likelihood function;
[0015] In the iterative solution, the current value of the normalization factor is kept known, and the MAP iterative reconstruction algorithm is used to obtain the latest estimate of the radioactivity distribution x. Then, while keeping the latest estimate of the radioactivity distribution x as known, the gradient descent method is used to obtain the latest estimate of the normalization factor.
[0016] The method of this embodiment can realize normalization correction of detection data of various radiation sources. In the process of acquiring the detection data, it is not necessary or necessary to specially set up a uniform central source, which greatly facilitates practical operations, such as using clinical scan data.
[0017] Optionally, the normalization correction factor includes: a geometric normalization factor vector f and an intrinsic detection efficiency vector ε of the PET system;
[0018] The objective function Φ(yx,f,ε) is:
[0019] Φ(yx,f,ε)=L(yx,f,ε)-β·R(x); Formula (A1)
[0020] R(x) is the penalty function; β is a non-negative weight factor; L(yx,f,ε) is the log-likelihood function of the PET detection data;
[0021] The log-likelihood function L(yx,f,ε) of PET detection data is:
[0022]
[0023] Among them, the PET radioactivity distribution x=[x1,x2,…,x M ] T and the intrinsic detection efficiency vector ε = [ε1,ε2,…,ε N ] T, geometric normalization factor vector f = [f 11 ,f 12 ,…,f NN ] T , y=[y 11 ,y 12 ,…,y NN ] T Indicates the actual detected data, A=[A ikj ] is the system matrix, j is 1…M, and is the spatial point source position in the PET system; y ik represents the photon pair events collected by detector i and detector k; p r refers to the Poisson probability distribution; ε i and ε k It is expressed as the intrinsic detection efficiency of detector i and detector k; f ik is the geometric normalization factor.
[0024] Optionally, the S30 includes:
[0025] S31, based on the objective function, using formula (A3) to obtain the maximum likelihood estimate of the radioactivity distribution x;
[0026]
[0027] S32. For the geometric normalization factor vector f, maximize the log-likelihood function L(yx,f,ε) to obtain the geometric normalization factor f ik The maximum likelihood estimate of is as shown in formula (A4);
[0028]
[0029] r ik represents the average value of random noise and scattered noise of detector i and detector k;
[0030] S33. For the intrinsic detection efficiency vector ε, maximize the log-likelihood function L(yx,f,ε) to obtain the intrinsic detection efficiency ε i The maximum likelihood estimate of is as shown in formula (A5);
[0031]
[0032] S34, based on the known initial values of the geometric normalization factor vector f and the intrinsic detection efficiency vector ε, when the radioactivity distribution x is unknown, the MAP iterative reconstruction algorithm is used to maximize the objective function Φ(yx,f,ε) to obtain the latest estimation result of the radioactivity distribution x;
[0033] Based on the latest estimated result of the known radioactivity distribution x and the initial value of the intrinsic detection efficiency ε, when the geometric normalization factor vector f is unknown, the geometric normalization factor is iteratively solved and the objective function Φ(yx,f,ε) is maximized to optimize and obtain the latest estimated result of the geometric normalization factor vector f;
[0034] Based on the latest estimated result of the known radioactivity distribution x and the latest estimated result of the geometric normalization factor vector f, the intrinsic detection efficiency is iteratively solved and the objective function Φ(yx,f,ε) is maximized, and the latest estimated result of the intrinsic detection efficiency vector ε is optimized;
[0035] The above alternating method is used to iterate so that the f in the geometric normalization factor vector f ik Converges to the geometric normalization factor f ik The maximum likelihood estimate of , the ε in the intrinsic detection efficiency vector ε i Converges to the intrinsic detection efficiency ε i The maximum likelihood estimate of .
[0036] Optionally, based on the latest estimated result of the known radioactivity distribution x and the initial value of the intrinsic detection efficiency vector ε, when the geometric normalization factor vector f is unknown, the gradient descent method and the first search step are iteratively processed and the objective function Φ(yx,f,ε) is maximized to optimize and obtain the latest estimated result of the geometric normalization factor vector f;
[0037] Based on the latest estimated results of the known radioactivity distribution x and the latest estimated results of the geometric normalization factor vector f, the gradient descent method and the second search step are iteratively processed and the objective function Φ(yx,f,ε) is maximized to optimize the latest estimated result of the intrinsic detection efficiency vector ε.
[0038] Optionally, the S34 includes:
[0039] The radioactivity distribution x is the result of reconstruction by other algorithms. In each iteration, the radioactivity distribution x is a known fixed value, and the geometric normalization factor and the intrinsic detection efficiency are iterated alternately;
[0040] or,
[0041] The intrinsic detection efficiency vector ε is a specified value. In each iteration, the intrinsic detection efficiency ε is a known fixed value, and the geometric normalization factor f and the radioactivity distribution x are iterated alternately;
[0042] Alternatively, the geometric normalization factor vector f is a specified value, and the geometric normalization factor is a known fixed value in each iteration, and the intrinsic detection efficiency and the radioactivity distribution x are alternately iterated.
[0043] In this embodiment, the order of the three variables is not fixed, and they can be freely combined and adjusted according to actual needs.
[0044] Optionally, the S30 further includes:
[0045] S35, normalizing the iteratively converged geometric normalization factor vector f according to the following formula (A6) to obtain the geometric normalization correction factor f for PET image reconstruction: ik ;
[0046] The iteratively converged intrinsic detection efficiency ε is normalized according to the following formula (A6) to obtain the intrinsic detection efficiency ε for PET image reconstruction: i ;
[0047]
[0048] Optionally, S34 includes:
[0049] Geometric normalization correction factor f ik The iterative formula is formula (A7):
[0050]
[0051] Search step length m represents the number of iterations;
[0052] Intrinsic detection efficiency ε i The iterative formula is formula (A8):
[0053]
[0054] The search step length n represents the number of iterations.
[0055] Optionally, the pre-given initial value of the normalized correction factor used to reconstruct the PET image is a parameter value used by the PET system;
[0056] If the PET detection data used for PET image reconstruction is the data of the reconstructed image, the radioactivity distribution x of the reconstructed image is used as the initial value of the radioactivity distribution x;
[0057] If the PET detection data used for PET image reconstruction is data of an unreconstructed image, the initial value of the radioactivity distribution x is a given value. It is understandable that a uniform value is usually selected for the entire space, for example, the initial value of all pixels is set to 1000.
[0058] In a second aspect, an embodiment of the present invention further provides a PET image reconstruction method, which includes:
[0059] acquiring PET detection data for PET image reconstruction;
[0060] Adopting any one of the methods for obtaining normalized correction factors for PET image reconstruction described in the first aspect above to obtain normalized correction factors for PET image reconstruction;
[0061] Based on the normalized correction factor and the PET detection data, the radioactivity distribution x of the corresponding PET image is reconstructed.
[0062] In a third aspect, an embodiment of the present invention further provides a PET system, comprising: a memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically executing the PET image reconstruction method described in the second aspect above.
[0063] (III) Beneficial effects
[0064] The method of the embodiment of the present invention performs joint maximum likelihood estimation on the PET radioactivity image and the normalized correction factor simultaneously during the reconstruction process, iteratively extracts the normalized correction factor, and applies it in real time to the reconstruction of the PET radioactivity activity distribution image.
[0065] The present invention satisfies the maximum likelihood estimation of PET radioactivity distribution while extracting the normalization factor, and the normalization correction process is integrated with the image reconstruction process. The normalization factor and PET radioactivity distribution are estimated simultaneously by using the maximum likelihood reconstruction process, which reduces the variance and reduces the influence of the algorithm error on the image reconstruction, and is conducive to improving the uniformity of image quality. That is, the influence of the traditional normalization algorithm error is reduced, which is conducive to improving the uniformity of image quality. The process does not require additional source cost and does not require complex mechanical positioning equipment.
[0066] Compared with the normalization correction method of the prior art, it does not require additional source costs, nor does it require complex mechanical positioning devices, and is applicable to various PET detection systems, such as long-axis PET systems. The method of the embodiment of the present invention has no restrictions on the radiation source, and no longer requires a uniform radiation source. The method of the embodiment of the present invention can also use conventional scanned patient data for normalization correction, and does not limit the type of PET detection data and other requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 A schematic flow chart of a method for obtaining a normalized correction factor for PET image reconstruction provided by an embodiment of the present invention;
[0068] Figure 2 A schematic diagram of a PET radioactive reconstructed image obtained by performing joint maximum likelihood estimation using the method of the present invention;
[0069] Figure 3 A schematic diagram of intrinsic detection efficiency obtained by performing joint maximum likelihood estimation using the method of the present invention;
[0070] Figure 4 The figure is a schematic diagram of a geometric normalization factor obtained by performing joint maximum likelihood estimation using the method of the present invention. DETAILED DESCRIPTION
[0071] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.
[0072] The purpose of the embodiment of the present invention is to optimize the normalization correction factor of the PET system, and the normalization factor is independent of the flight time measurement result, and the flight time dimension is not considered in the following descriptions.
[0073] In addition, the normalized correction factor acquisition method of the embodiment of the present invention performs joint maximum likelihood estimation on the PET radioactive image and the normalized correction factor at the same time during the reconstruction process, iteratively extracts the normalized correction factor, and applies it in real time in the reconstruction of the PET radioactive image, thereby saving the algorithm processing process and improving the quality of the reconstructed image. The italics and boldface in the following embodiments represent vectors. In addition, in some embodiments, estimated values are used, and in some embodiments, estimated results are used, and their meanings are consistent, and PET image reconstruction is the reconstruction of the PET radioactive activity distribution image.
[0074] Embodiment 1
[0075] like Figure 1 As shown, an embodiment of the present invention provides a method for obtaining a normalized correction factor for PET image reconstruction. The execution subject of the method of this embodiment may be a control device / electronic device for the method for obtaining a normalized correction factor for PET image reconstruction. The control device may be integrated in an acquisition device of a PET system or in a separate computer processing device. The method for obtaining a normalized correction factor includes the following steps:
[0076] S10, acquiring PET detection data for PET image reconstruction and a predetermined normalization correction factor for PET image reconstruction and an initial value of radioactivity distribution x;
[0077] In this step, the PET detection data is data collected by a detection device of a uniform radiation source, a non-uniform radiation source, or a non-centrally positioned radiation source. Generally, the PET detection data in the prior art must be limited to data collected by a detection device of a uniform radiation source, but in this embodiment, data collected by a detection device of a non-uniform radiation source can be selected, so that the PET image reconstructed by the method of this embodiment has high quality, meets the requirements, and can achieve uniformity of the reconstructed image.
[0078] S20, constructing a target function Φ(yx,f,ε) with unknowns being a normalization correction factor (such as a geometric normalization factor f and an intrinsic detection efficiency ε of the PET system) and a radioactivity distribution x according to the PET detection data and based on the principle of Poisson distribution, wherein the target function is composed of a log-likelihood function with the normalization correction factor and the radioactivity distribution x as unknowns and a penalty function R(x) with the radioactivity distribution x as an unknown;
[0079] S30, iteratively solving the objective function Φ(yx,f,ε) according to the normalized correction factor and the initial value of the radioactivity distribution x, so that the normalized correction factor converges to the maximum likelihood estimate, and obtaining the iterative normalized correction factor as the normalized correction factor for PET image reconstruction;
[0080] Wherein, the maximum likelihood estimate is a value obtained by performing maximum likelihood estimation on the following log-likelihood function;
[0081] In the iterative solution, the current value of the normalization factor is kept known, and the MAP iterative reconstruction algorithm is used to obtain the latest estimate of the radioactivity distribution x. Then, while keeping the latest estimate of the radioactivity distribution x as known, the gradient descent method is used to obtain the latest estimate of the normalization factor.
[0082] Compared with the normalization correction method of the prior art, the method of this embodiment satisfies the maximum likelihood estimation of PET radioactivity distribution while extracting the normalization factor. The normalization correction process can be integrated with the image reconstruction process, and the normalization factor is adjusted using the maximum likelihood reconstruction process to reduce the impact of the error of the variance reduction algorithm, which is beneficial to improving the uniformity of image quality.
[0083] The normalization correction method of this embodiment has no restrictions on the radiation source and does not require a uniform radiation source. Normalization correction can be performed using patient data from conventional scans. It does not require additional source costs or complex mechanical alignment equipment, and is applicable to various PET detection systems, such as long-axis PET systems.
[0084] Embodiment 2
[0085] The normalized correction factor acquisition method of this embodiment performs joint maximum likelihood estimation on the PET radioactive image and the normalized correction factor simultaneously during the reconstruction process, iteratively extracts the normalized correction factor, and applies it in real time in PET image reconstruction.
[0086] To better understand the method of Example 1, each step and principle is described in detail below:
[0087] The first step is to obtain PET detection data for PET image reconstruction.
[0088] Step 2: The PET acquisition process can be modeled as follows:
[0089]
[0090] The meaning of each parameter in the formula is as follows:
[0091] Represents the average value of the detected data. Since PET detects photon pairs generated by positron annihilation, the two numbers in the subscript represent the numbers of the two detector crystals corresponding to the detected photon pairs, and N represents the number of detector crystals. A photon pair event detected by PET corresponds to two detectors. For example, detectors 1 and 5 define an event, and detectors 10 and 2 define an event, etc. At this time, it is equivalent to the permutation and combination of the detectors themselves.
[0092] x=[x1,x2,…,x M ] T represents the unknown PET radioactivity distribution, and M represents the size of the radioactivity distribution image space.
[0093] A=[A ikj ] is the system matrix, which expresses the probability of spatial point source j being detected by detector i and detector k simultaneously in the PET system in mathematical form, including the geometric detection efficiency factor, photon detection point spread factor and photon attenuation factor in the photon pair detection process. The geometric detection efficiency is expressed as the probability of a photon pair reaching the surface of the detector pair, which depends on the change of the geometric solid angle of the detector pair relative to each pixel position; the photon point spread function is expressed as the response function of the detector crystal to the incident photon; and the photon attenuation is expressed as the probability of a photon pair attenuating before being detected.
[0094] is the normalized correction factor matrix, which is used to correct the uneven detection efficiency of the detector crystal. i, k represent detector i and detector k, and each pair of detectors corresponds to a normalized correction factor. r = [r 11 ,r 12 ,…,r NN ] T Represents the average value of random noise and scattered noise.
[0095] In addition, in order to reduce the statistical data requirements, the normalization correction parameter evaluation usually adopts a model-based method. The normalization correction parameter can be modeled as:
[0096]
[0097] where εi and ε k It is expressed as the intrinsic detection efficiency of detector i and detector k, and is used to correct the inconsistency of detection efficiency caused by the individual luminous efficiency of the crystal, crystal packaging, coupling between the crystal and the photomultiplier tube (PMT) or silicon photomultiplier tube (SiPM), electronic system, etc. Usually, the detection equipment will be normalized and calibrated before leaving the factory to obtain the detection efficiency of the system, but the detection efficiency will change as the system works, so it needs to be updated regularly.
[0098] Geometric normalization factor f ik It reflects that the detection efficiency of the photon pairs detected by detectors i and k is not uniform due to the influence of the geometric factors of the PET system, which depends on the positions of the two detectors, the incident angle of the detector line relative to the detector plane (i.e., the actual surface of the detector), and the distance between the detector pair line and the center of the scanner field of view. It is understandable that for the PET system, the normalization correction is to correct the influence of the above-mentioned detector detection efficiency and system geometric factors on the collected data. In addition, the correction factors related to the counting rate (dead time correction factor, pile-up effect correction factor) are determined by other experiments and are not involved in the embodiments of the present invention. That is, the intrinsic detection efficiency and geometric normalization factor to be solved have nothing to do with the activity of the radiation source, that is, they have nothing to do with the counting rate collected in real time.
[0099] Step 3: Construct the objective function with unknowns.
[0100] 1) PET detection data follows Poisson distribution, and the unknown number is the PET radioactivity distribution x = [x1, x2, …, x M ] T and the intrinsic detection efficiency ε=[ε1,ε2,…,ε N ] T , geometric normalization factor f = [f 11 ,f 12 ,…,f NN ] T , then the log-likelihood function of the detection data is expressed as:
[0101]
[0102] In formula (3) y=[y 11 ,y 12 ,…,y NN ] T Indicates the data actually detected.
[0103] 2) Substitute formula (1) into formula (3), ignore the terms unrelated to the unknown number, and the log-likelihood function can be written as:
[0104]
[0105] 3) In order to improve the data ill-conditioning problem in the process of solving the log-likelihood function, the objective function can be adjusted with the help of an explicit regularization process, that is, a penalty function R(x) is added to the log-likelihood function to selectively penalize specific undesirable features, and the maximum likelihood solution with this penalty term is calculated.
[0106] The corresponding objective function is:
[0107] Φ(yx,f,ε)=L(yx,f,ε)-β·R(x)(5)
[0108] β is a non-negative weight factor used to weigh the importance of the log-likelihood function and the penalty function.
[0109] Step 4: For the objective function, determine the iterative convergence information and iterative formula of each unknown number.
[0110] 1) When iteratively solving the unknowns of the objective function, such as the PET radioactivity distribution x, the regularized maximum likelihood estimate can be calculated using the maximum a posteriori probability algorithm MAP (Maximum aposteriori):
[0111]
[0112] Of course, if the parameter β is 0, then formula (6) becomes the traditional maximum likelihood maximum expectation value algorithm MLEM (Maximum Likelihood Expectation Maximization) or its accelerated algorithm ordered subset maximum expectation value algorithm OSEM (Ordered Subset Expectation Maximization) for iterative solution.
[0113] 2) The normalized correction factor is composed of the geometric normalization factor vector f and the intrinsic detection efficiency vector ε, which is represented as the normalized calibration parameter of the detection data y. The objective function to be optimized is the log-likelihood function L(yx,f,ε).
[0114] The scalar penalty function R(x) is only related to the PET radioactivity distribution x and has nothing to do with the normalized correction factor, so it can be ignored when the normalized correction factor is derived.
[0115] 2-1) First, maximize the log-likelihood function (4) for the geometric normalization factor vector f to obtain f ik The maximum likelihood estimate of :
[0116]
[0117] Both scattered photons and random photons are incident on the detector surface at a wide range of angles. The angle of incidence of the photons is not restricted by the detector position, which is different from the case of real photon pairs. Therefore, any geometric factors of scattered and random events are averaged over a large range of angles and can be effectively ignored, so r ik / f ik ≈r ik Therefore, ik The maximum likelihood estimate of can be approximated as:
[0118]
[0119] 2-2) Secondly, the gradient descent method is selected as the optimization algorithm for iteratively solving the geometric normalization factor vector f, and the maximum likelihood solution of the objective function is recursively approximated along the gradient descent direction. The iterative formula can be expressed as:
[0120]
[0121] In the formula, α represents the search step size in the gradient direction, and m represents the number of iterations. Since the objective function L(yx,f,ε) is known, it is necessary to solve the step size variable α of the iterative function. Transform formula (8):
[0122]
[0123] Comparing formula (10) with formula (9), we can get the iterative formula:
[0124]
[0125] The search step length
[0126] According to formula (11), when the iteration converges, the geometric normalization factor f ik converges to its maximum likelihood estimate (8).
[0127] 2-3) Next, the log-likelihood function (4) is maximized for the intrinsic detection efficiency vector ε to obtain ε i The maximum likelihood estimate of :
[0128]
[0129] 2-4) Further, the gradient descent method is selected as the optimization algorithm for iteratively solving the intrinsic detection efficiency vector ε, and the maximum likelihood solution of the objective function is recursively approached along the gradient descent direction, ε i The iteration formula can be expressed as:
[0130]
[0131] In the formula, η represents the search step in the gradient direction, and n represents the number of iterations. Since the objective function L(yx,f,ε) is known, it is necessary to solve the step variable η of the iterative function. Transform formula (12)
[0132]
[0133] Comparing formula (14) with formula (13), we can get the iterative formula:
[0134]
[0135] The search step length
[0136] According to formula (15), when the iteration converges, the intrinsic detection efficiency ε i converges to its maximum likelihood estimate (12).
[0137] Step 5: Based on the above objective function, the iterative convergence information of each unknown number and the iterative formula are iteratively solved.
[0138] In order to obtain the estimated values of the unknown PET radioactivity distribution x and the vectors f and ε, it is necessary to optimize the objective function (5) of their joint distribution. Since the above objective function is a very complex function for the unknowns x, f, and ε, it is difficult to obtain an analytical solution for formula (5), so it is necessary to use an iterative algorithm to gradually approach the optimal solution. For the feasibility of the algorithm implementation, it is necessary to simplify the optimization process and explain it as follows:
[0139] First, the vectors f and ε are kept constant, and the objective function is maximized for the PET radioactivity distribution x, using the traditional MAP iterative reconstruction algorithm;
[0140] Then, the PET radioactivity distribution x and the intrinsic detection efficiency vector ε are kept constant, and the objective function is maximized for the geometric normalization factor vector f, using the iterative formula (11);
[0141] Finally, we choose to keep the PET radioactivity distribution x and the geometric normalization factor vector f as constants, maximize the objective function with respect to the intrinsic detection efficiency vector ε, and adopt the iterative formula (15).
[0142] As described above, the maximization operation is performed alternately, and the normalization correction parameters are continuously modified to make them close to the actual normalization situation, and finally the estimated values of x, f, and ε that meet the requirements of the maximization objective function are obtained.
[0143] It should be emphasized that the iteration order of the three unknowns is not limited in this patent. It is also possible to iteratively solve the PET radioactivity distribution x first, then iteratively solve the intrinsic detection efficiency vector ε, and finally iteratively solve the geometric normalization factor vector f. It is also possible to choose to use the system default result for the intrinsic detection efficiency vector ε without solving it, and only alternately iterate the solution of the PET radioactivity distribution x and the geometric normalization factor vector f. It is also possible to choose to use the result x of conventional reconstruction without solving it, and only alternately iterate the solution of the intrinsic detection efficiency vector ε and the geometric normalization factor vector f.
[0144] It should be noted that the iterative initial value of the geometric normalization factor vector f or the intrinsic detection efficiency vector ε can select the parameter result previously used by the system, or all parameter values can be selected as 1.
[0145] The geometric normalization factor vector f or the intrinsic detection efficiency vector ε is iteratively solved for the maximum likelihood value. In the solution process, there is no requirement for the source of the radioactivity distribution, that is, any radioactive object can be used for normalization correction and parameter optimization, such as patient scanning data.
[0146] Step 6: Iterate the intrinsic detection efficiency ε solved in step 5 i and the geometric normalization factor f ik A normalization process is performed to obtain a normalized correction factor for PET image reconstruction.
[0147] In formula (1) and formula (2), if the geometric normalization factor f or the intrinsic detection efficiency ε is multiplied by a constant, then the reconstructed PET radioactivity distribution x divided by the same constant can still obtain the same measured average value. This means that the normalization correction factor estimated by the above algorithm may be a constant multiple of the actual normalization correction factor, and this constant cannot be determined by the algorithm itself. However, since the normalization correction factor corrects the relative change in detection efficiency, its mean value can be set to 1, and this information can be used to correct the calculated geometric normalization factor f ik Or intrinsic detection efficiency ε i Normalize, that is:
[0148]
[0149] You can choose to update the variable after each iteration or only update it at the last iteration, which is not limited in this embodiment.
[0150] Step 7: Based on the normalized correction factor reconstructed from the PET image, the PET image of the PET radioactivity distribution x is obtained.
[0151] like Figure 2 As shown, Figure 2FIG. 4 is a schematic diagram showing a PET radioactive reconstructed image obtained by performing joint maximum likelihood estimation using the above method, Figure 3 Schematic diagram of intrinsic detection efficiency obtained by joint maximum likelihood estimation using the above method, Figure 4 Schematic diagram of the geometric normalization factor obtained by joint maximum likelihood estimation using the above method.
[0152] Embodiment 3
[0153] This embodiment provides a PET image reconstruction method, which includes:
[0154] acquiring PET detection data for PET image reconstruction;
[0155] Adopting the method for obtaining a normalized correction factor for PET image reconstruction described in the first embodiment or any one of the embodiments above to obtain a normalized correction factor for PET image reconstruction;
[0156] Based on the normalized correction factor and the PET detection data, the radioactivity distribution x of the corresponding PET image is reconstructed.
[0157] In addition, an embodiment of the present invention further provides a PET system, which includes: a memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory to specifically perform the above-mentioned PET image reconstruction method.
[0158] Generally, the above-mentioned PET image reconstruction method can be specifically executed by a control device in a PET system.
[0159] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.
[0160] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0161] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.
[0162] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.
Claims
1. A method for obtaining a normalized correction factor for PET image reconstruction, characterized in that: include: S10, acquiring PET detection data y for PET image reconstruction and a predetermined normalized correction factor for PET image reconstruction and an initial value of radioactivity distribution x; the PET detection data is data collected by a detection device of a uniform radiation source, a non-uniform radiation source or a non-centrally positioned radiation source; S20, constructing an objective function with unknowns being a normalized correction factor and a radioactivity distribution x according to the PET detection data and based on the principle of Poisson distribution, wherein the objective function is composed of a log-likelihood function with the normalized correction factor and the radioactivity distribution x as unknowns and a penalty function R(x) with the radioactivity distribution x as an unknown; S30, iteratively solving the objective function according to the normalized correction factor and the initial value of the radioactivity distribution x, so that the normalized correction factor converges to the maximum likelihood estimation value, and obtaining the iterative normalized correction factor as the normalized correction factor for PET image reconstruction; Wherein, the maximum likelihood estimate is a value obtained by performing maximum likelihood estimation on the log-likelihood function; In the iterative solution, the current value of the normalization factor is kept known, and the MAP iterative reconstruction algorithm is used to obtain the latest estimate of the radioactivity distribution x. Then, while keeping the latest estimate of the radioactivity distribution x as known, the gradient descent method is used to obtain the latest estimate of the normalization factor.
2. The method for obtaining a normalized correction factor according to claim 1, characterized in that: The normalization correction factor includes: a geometric normalization factor vector f and an intrinsic detection efficiency vector ε of the PET system; The objective function Φ(yx,f,ε) is: Φ(yx,f,ε)=L(yx,f,ε)-β·R(x); Formula (A1) R(x) is the penalty function; β is a non-negative weight factor; L(yx,f,ε) is the log-likelihood function of the PET detection data; The log-likelihood function L(yx,f,ε) of PET detection data is: Among them, the PET radioactivity distribution x=[x1,x2,…,x M ] T and the intrinsic detection efficiency vector ε = [ε1,ε2,…,ε N ] T , geometric normalization factor vector f = [f 11 ,f 12 ,…,f NN ] T , y=[y 11 ,y 12 ,…,y NN ] T Indicates the actual detected data, A=[A ikj ] is the system matrix, j is 1…M, and is the spatial point source position in the PET system; y ik represents the photon pair events collected by detector i and detector k; p r refers to the Poisson probability distribution; ε i and εk represent the intrinsic detection efficiency of detector i and detector k; f ik is the geometric normalization factor.
3. The method for obtaining a normalized correction factor according to claim 2, characterized in that: The S30 includes: S31, based on the objective function, using formula (A3) to obtain the maximum likelihood estimate of the radioactivity distribution x; S32. For the geometric normalization factor vector f, maximize the log-likelihood function L(yx,f,ε) to obtain the geometric normalization factor f ik The maximum likelihood estimate of is as shown in formula (A4); r ik represents the average value of random noise and scattered noise of detector i and detector k; S33. For the intrinsic detection efficiency vector ε, maximize the log-likelihood function L(yx,f,ε) to obtain the intrinsic detection efficiency ε i The maximum likelihood estimate of is as shown in formula (A5); S34, based on the known initial values of the geometric normalization factor vector f and the intrinsic detection efficiency vector ε, when the radioactivity distribution x is unknown, the MAP iterative reconstruction algorithm is used to maximize the objective function Φ(yx,f,ε) to obtain the latest estimation result of the radioactivity distribution x; Based on the latest estimated result of the known radioactivity distribution x and the initial value of the intrinsic detection efficiency vector ε, when the geometric normalization factor vector f is unknown, the geometric normalization factor is iteratively solved and the objective function Φ(yx,f,ε) is maximized to optimize and obtain the latest estimated result of the geometric normalization factor vector f; Based on the latest estimated result of the known radioactivity distribution x and the latest estimated result of the geometric normalization factor vector f, the intrinsic detection efficiency is iteratively solved and the objective function Φ(yx,f,ε) is maximized, and the latest estimated result of the intrinsic detection efficiency vector ε is optimized; The above alternating method is used to iterate so that the f in the geometric normalization factor vector f ik Converges to the geometric normalization factor f ik The maximum likelihood estimate of , the ε in the intrinsic detection efficiency vector ε i Converges to the intrinsic detection efficiency ε i The maximum likelihood estimate of .
4. The method for obtaining a normalized correction factor according to claim 3, characterized in that: Based on the latest estimated result of the known radioactivity distribution x and the initial value of the intrinsic detection efficiency vector ε, when the geometric normalization factor vector f is unknown, the gradient descent method and the first search step are used to iteratively process and maximize the objective function Φ(yx,f,ε) to optimize and obtain the latest estimated result of the geometric normalization factor vector f; Based on the latest estimated results of the known radioactivity distribution x and the latest estimated results of the geometric normalization factor vector f, the gradient descent method and the second search step are iteratively processed and the objective function Φ(yx,f,ε) is maximized to optimize the latest estimated result of the intrinsic detection efficiency vector ε.
5. The method for obtaining a normalized correction factor according to claim 3 or 4, characterized in that: The S34 includes: The radioactivity distribution x is the result of reconstruction by other algorithms. In each iteration, the radioactivity distribution x is a known fixed value, and the geometric normalization factor and the intrinsic detection efficiency are iterated alternately; or, The intrinsic detection efficiency vector ε is a specified value. In each iteration, the intrinsic detection efficiency vector ε is a known fixed value, and the geometric normalization factor and the radioactivity distribution x are alternately iterated.
6. The method for obtaining a normalized correction factor according to claim 3 or 4, characterized in that: The S30 further includes: S35, normalizing the iteratively converged geometric normalization factor vector f according to the following formula (A6) to obtain the geometric normalization correction factor f for PET image reconstruction: ik ; The iteratively converged intrinsic detection efficiency vector ε is normalized according to the following formula (A6) to obtain the intrinsic detection efficiency ε for PET image reconstruction: i ; 7. The method for obtaining a normalized correction factor according to claim 3 or 4, characterized in that: S34 includes: Geometric normalization correction factor f ik The iterative formula is formula (A7): Search step length m represents the number of iterations; Intrinsic detection efficiency ε i The iterative formula is formula (A8): The search step length n represents the number of iterations.
8. The method for obtaining a normalized correction factor according to any one of claims 1 to 4, characterized in that: The predetermined initial value of the normalized correction factor used to reconstruct the PET image is a parameter value used by the PET system; If the PET detection data used for PET image reconstruction is the data of the reconstructed image, the radioactivity distribution x of the reconstructed image is used as the initial value of the radioactivity distribution x; If the PET detection data used for PET image reconstruction is data of an unreconstructed image, the initial value of the radioactivity distribution x is a given value.
9. A PET image reconstruction method, characterized in that: include: acquiring PET detection data for PET image reconstruction; Adopting the method for obtaining a normalized correction factor for PET image reconstruction according to any one of claims 1 to 8 above to obtain a normalized correction factor for PET image reconstruction; Based on the normalized correction factor and the PET detection data, the radioactivity distribution x of the corresponding PET image is reconstructed.
10. A PET system, characterized in that: include: A memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically executing the PET image reconstruction method described in claim 9.
Citation Information
Patent Citations
Test method as well as correction method and device for detection efficiency of PET (positron emission tomography) system
CN106264588A
Correction information acquisition method for performing attenuation correction on PET activity distribution image
CN109978966A
PET image reconstruction method and system
CN112529977A
Image-based point-spread-function modelling in time-of-flight positron-emission-tomography iterative list-mode reconstruction
US20150199302A1
