Systems, methods and non-transitory computer-readable media for imaging

Abstract

A non-transitory computer readable medium storing instructions that can be read and executed by a workstation (18), the workstation comprising at least one electronic processor (20) to perform an imaging method (100). The method comprises: receiving imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps the volume (k) along the axial direction; and generating an image of the volume (k) using an iterative image reconstruction process, wherein iterations of the iterative image reconstruction process include: calculating a local penalty function for suppressing noise on the volume (k), including reducing a value of the local penalty function in an overlapping region; generating an updated image of the volume (k) using imaging data from the volume (k) and also using the local penalty function.

Description

Technical Field

[0001] The following generally relates to medical imaging techniques, positron emission tomography (PET) imaging and image reconstruction techniques, regularized iterative image reconstruction techniques, and related techniques. Background Art

[0002] Image quality in positron emission tomography (PET) is strongly dominated by noise. Iterative maximum a posteriori penalized reconstruction (PR) using edge-preserving or contrast-preserving priors is a method to suppress noise while preserving small features (e.g., tumors) that may be clinically significant. Some edge-preserving or contrast-preserving penalties include: (1) relative difference penalty (RDP) (e.g., see J. Nuyts et al., "A concave prior penalizing relative differences for maximum-a-posteriori reconstruction in emission tomography" IEEE Trans. Nucl. Sci., 2002); (2) anisotropic diffusion penalty (ADP) (e.g., see H. Zhu, "Image reconstruction for positron emission tomography using fuzzy nonlinear anisotropic diffusion penalty", Med Bio Eng Comput, 2006), etc. In most edge-preserving penalties, tuning parameters establish the sensitivity to edge transitions in the reconstructed image, and imaging data below the sensitivity threshold are smoothed by a strong low-pass filter.

[0003] Commercial PET cameras have a limited axial field of view (FOV) due to the limited axial extent of the PET detector ring or array. Therefore, a "whole body" patient scan is typically acquired in a series of "step-and-shoot" modes at multiple bed positions, with a certain overlap between bed positions. For practical reasons, each bed position is typically reconstructed separately without waiting for the PET data for the next bed position to be fully acquired, and the reconstructed individual frame images are woven into a single whole body image. Each bed position overlaps to ensure smooth signal transitions in the axial direction. Typically, for whole body studies, more than 50% of the scanned axial extent will be covered in overlapping slices. Therefore, in each reconstructed image volume, there will be one or two sub-volumes covered by counts acquired from the previous bed position or the next bed position (note, as discussed in this article, it is assumed that the number of bed positions considered in all cases is at least greater than 1).

[0004] Instead of reconstructing each bed position separately and then weaving the images together in image space, in an alternative approach, the counts are pre-combined (in sinogram format or list mode) before reconstruction without overly complicating the imaging workflow (see, e.g., Z. Sun et al., “Reconstruction and combination of PET multi-bed image,” U.S. Publication No. US 2017 / 0103551). However, previous work has shown that at least the ordered subset expectation maximization (OSEM) iterative image reconstruction algorithm exhibits quasi-linear behavior and that there is little practical benefit to pre-combining counts before reconstruction, except for ultra-low count studies (see, e.g., S. Ross et al., “A method of overlap correction forfully 3D OSEM reconstruction of PET data,” MIC, 2004). With each bed position reconstructed separately, each bed position image can have a noisier appearance in edge slices due to the effective “loss” of data at the edges. However, the noise can generally be compensated during the image weaving process (i.e., when combining the individual bed position images into a single whole-body image).

[0005] Disclosed below are new and improved systems and methods that address these problems. Summary of the invention

[0006] In one disclosed aspect, a non-transitory computer-readable medium stores instructions that can be read and executed by a workstation that includes at least one electronic processor to perform an imaging method. The method includes: receiving imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps the volume (k) along the axial direction; and using an iterative image reconstruction process to generate an image of the volume (k), wherein iterations of the iterative image reconstruction process include: calculating a local penalty function for suppressing noise on the volume (k), including reducing a value of the local penalty function in an overlapping region; using imaging data from the volume (k) and also using the local penalty function to generate an updated image of the volume (k).

[0007] In another disclosed aspect, an imaging system includes: a positron emission tomography (PET) imaging device. At least one electronic processor is programmed to: receive imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps the volume (k) along the axial direction; and generate an image of the volume (k) using an iterative image reconstruction process, wherein iterations of the iterative image reconstruction process include: calculating a local penalty function for suppressing noise on the volume (k), including reducing a value of the local penalty function in an overlap region, the local penalty function being dependent on a total coincidence count through the penalized voxel, and the value of the local penalty function being reduced in the overlap region by an amount that compensates for the total coincidence count through the penalized voxel in the subsequent volume k+1; and generating an updated image of the volume (k) using imaging data from the volume (k) and also using the local penalty function. The reconstruction includes: combining the number of counts of the volume (k) from the imaging data into a total volume; and reconstructing the total volume into a reconstructed image.

[0008] In another disclosed aspect, an imaging method performed on imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps the volume (k) along the axial direction, the method comprising: generating an image of the volume (k) using an iterative image reconstruction process, wherein iterations of the iterative image reconstruction process include: calculating a local penalty function for suppressing noise on the volume (k), including reducing a value of the local penalty function in an overlapping region, the local penalty function being dependent on a local geometric sensitivity of the PET imaging device, and the value of the local penalty function being reduced in the overlapping region by an amount compensating for a loss in geometric sensitivity due to an axial truncation of the volume (k); generating an updated image of the volume (k) using imaging data from the volume (k) and also using the local penalty function. The reconstruction includes: reconstructing the volume (k) of the imaging data into a reconstructed image separately from other volumes (k) of the imaging data.

[0009] In another disclosed aspect, an imaging method is performed on imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps with the volume (k) along the axial direction, the method comprising: using an iterative image reconstruction process to generate an image of the volume (k), wherein iterations of the iterative image reconstruction process include: using two or more gate phases and motion detected in the gate phases to adjust the imaging data using respiratory gating; calculating a local penalty function for suppressing noise on the volume (k), including reducing the value of the local penalty function in the overlapping area; using imaging data from the volume (k) and also using the local penalty function to generate an updated image of the volume (k).

[0010] One advantage resides in utilizing all available counts from a PET imaging device to appropriately tune parameters of a reconstruction penalty function.

[0011] Another advantage resides in providing an edge-preserving penalized reconstruction process that avoids over-smoothing of the final image resulting in potential loss of image features.

[0012] Another advantage resides in providing an edge-preserving penalized reconstruction process that avoids under-smoothing of the final image resulting in potential noise in the image.

[0013] Another advantage is that when combining data in overlapping regions of imaging data prior to image reconstruction, the penalty function is reduced in proportion to the effective sensitivity increase introduced by combining adjacent images in the overlapping region to maintain the spatial resolution of the reconstructed image.

[0014] Another advantage resides in reducing the penalty function to preserve the spatial resolution of the reconstructed image when data in overlapping regions of imaging data are not combined prior to image reconstruction.

[0015] Another advantage resides in scaling the penalty function using the acquisition duration ratio between two imaging frames.

[0016] Another advantage resides in dynamically adjusting a penalty parameter of a penalty function using motion corrected respiratory or cardiac gating or dynamic studies adjusted with count endpoints.

[0017] A given embodiment may provide none of the aforementioned advantages, may provide one, two, more, or all of the aforementioned advantages, and / or may provide other advantages as will become apparent to one of ordinary skill in the art upon reading and understanding the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The disclosure may take the form of various components and arrangements of components, and may take the form of various steps and arrangements of steps.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the disclosure.

[0019] Figure 1 schematically illustrates an imaging system according to one aspect;

[0020] Figure 2 and Figure 3 Shows Figure 1 An exemplary flow chart operation of a system of; and

[0021] Figure 4 ac shows the standard reconstructed image and the Figure 2 Flowchart of the operation for comparison of reconstructed images. DETAILED DESCRIPTION

[0022] Iterative penalized PET image reconstruction employs an update function that includes a penalty term that is specified to penalize (i.e., suppress) noise while preserving edges. For example, the relative difference penalty (RDP) employs a penalty P∝1 / γ, where the term γ is an edge preservation parameter. Larger values ​​of γ produce smaller penalties P, which enhance the preservation of features (i.e., "edges") (e.g., small lesions), but also result in more noise being preserved; whereas smaller values ​​of γ increase the penalty, thereby improving noise suppression, but also result in the removal of clinically important structures (e.g., "edges") (e.g., small lesions). In variant methods, for example, by using The penalty is locally tuned, where λ i is the value of the voxel indexed by i, and s i is the geometric sensitivity at voxel i, which is proportional to the number of lines of response (LORs) passing through voxel i. This formula is premised on the expectation that if more LORs pass through voxel i (hence higher s i ), then there should be more counts nearby, and thus the noise at voxel i should be lower, so that edges can be enhanced with a smaller penalty P while still suppressing the (relatively low) noise. In an alternative tuning, the geometric sensitivity s is i is replaced by a term that quantifies the total counts in the local neighborhood around voxel i. Again, higher total counts in the local neighborhood tend to correspond to lower noise, so that a lower penalty P can be used.

[0023] It is recognized herein that such a penalty framework may be less effective in the case of overlapping regions of step-and-shoot PET imaging. In a step-and-shoot approach (also known as multi-bed imaging or similar nomenclature), a patient is stepped through a scanner to image successive axial portions of the patient ("bed positions"). Thus, the data acquired at each bed position is truncated axially at an edge slice corresponding to the location of the PET scanner's outermost detector ring at that bed position. Typically, the PET counts acquired at each bed position are reconstructed separately (but information from adjacent bed positions may be used to estimate out-of-FOV scatter), and the resulting bed position images are knitted together at the overlap by weighted averaging the voxel values ​​in the overlapping regions of the bed position images. However, in this approach, the geometric sensitivity s for that bed position is used. i Compute the penalized local tuning. For the outer slice of the bed position, the sensitivity s i becomes smaller because LORs ending outside the axial FOV of the bed position do not contribute to the sensitivity s i For extreme edge slices, only the LOR in the plane of the slice contributes to s i This means punishment The occlusions are larger at the outer slices, leading to excessive noise suppression in the overlap region and increasing the potential to suppress clinically relevant image features (e.g., small tumors). Subsequent knitting together of adjacent bed position images at the overlap cannot recover these suppressed features because both overlap regions will have this same “edge effect” in terms of penalty, leading to loss of features. i An alternative formulation that substitutes total counts in a local neighborhood would be similarly affected because counts for LORs that terminate outside the truncated axial FOV of the bed position are not acquired.

[0024] In some disclosed embodiments, the local penalty in the overlapping region is reduced to account for the above effects.

[0025] In some embodiments disclosed herein, if a truncated volume k is continued into an adjacent overlapping truncated volume k+1, a reduction proportional to the increase in effective sensitivity (or increase in effective total neighborhood count) will be achieved. In one approach using geometric sensitivity to adjust the penalty, this is equivalent to using replace Among them, s i,k+1 The term takes into account the part that is not part of volume k (and therefore does not contribute to the geometric sensitivity s i,k ) but is part of the overlapping volume k+1 (and thus contributes to the geometric sensitivity s i,k+1A similar formulation can be used if γ is adjusted locally using the total counts in the local neighborhood by taking the sum of the total counts in the neighborhood of voxel i from both overlapping volumes k and k+1.

[0026] In the above embodiment, each volume k is still reconstructed separately. In an alternative joint reconstruction embodiment, the PET counts from all volumes are first combined and then reconstructed together. In this case, again using Here, a suitable formulation could be Among them, λ i is the value of voxel i resulting from the (single) joint reconstruction of the combined datasets.

[0027] In a variant embodiment, if the acquisition time of each bed position is different, the sensitivity s i,k and i,k+1 scales proportionally with the acquisition time. This is because shorter acquisition times result in fewer counts being acquired and correspondingly lower sensitivity. Thus, if for corresponding overlapping bed positions k and k+1, the acquisition time is T k and T k+1 , then scale to and

[0028] In other embodiments disclosed herein, similar methods can be applied to motion compensated PET imaging using (e.g. respiratory) gating. In this case, consider two consecutive gate phases φ k and φ k+1 , where some motion occurs between the two phases, represented by the motion vector Δr. A possible formulation of the adjusted edge-preserving parameter is then

[0029] refer to Figure 1 , an illustrative medical imaging system 10 is shown. Figure 1 As shown, the system 10 includes an imaging or image acquisition device 12. In one example, the image acquisition device 12 may include a PET imaging device including a PET gantry 13 and an array of radiation detectors 14 (in Figure 115 ; typically, the radiation detectors of the PET gantry are arranged as a series of PET detector rings that are arranged to span the axial FOV). The illustrative image acquisition device 12 is a PET / CT scanner that also includes a computed tomography (CT) gantry 15. A patient table (or bed) 16 is arranged to load a patient into an examination region 17 of the imaging device 12, such as into a bore of the CT gantry 15 or into a bore of the PET gantry 13. Typically, the CT gantry 15 is used to acquire CT images of a subject, which are converted into an attenuation map of the subject. The PET gantry 13 acquires PET imaging data reconstructed from the CT images using the attenuation map to account for radiation absorption within the patient in the PET image reconstruction.

[0030] The system 10 also includes a computer or workstation or other electronic data processing device 18 having typical components, such as at least one electronic processor 20, at least one user input device (e.g., a mouse, keyboard, trackball, touchpad, and / or the like) 22, and a display device 24 (e.g., an LCD display, an OLED display, a plasma display, etc.). In some embodiments, the display device 24 may be a separate component from the computer 18, and / or may include two or more displays. The workstation 18 may also include one or more databases or non-transitory storage media 26 (e.g., a disk, RAID, or other magnetic storage media; a solid-state drive, a flash drive, an electrically erasable read-only memory (EEROM) or other electronic memory; an optical disk or other optical storage device; various combinations thereof; etc.). The display device 24 is configured to display images acquired by the imaging system 10, and typically also displays a graphical user interface (GUI) 28 including various user dialog boxes, for example, each having one or more fields, radial selection buttons, etc., to receive user input from the user input device 22.

[0031] The at least one electronic processor 20 is operatively connected to one or more databases 26 that store instructions readable and executable by the at least one electronic processor 20 to perform the disclosed operations, including performing the imaging method or process 100. In some examples, the imaging method or process 100 can be performed at least in part through cloud processing.

[0032] refer to Figure 2, an illustrative embodiment of an imaging method 100 is graphically shown as a flow chart. At 102, at least one electronic processor 20 is programmed to receive, otherwise operate or control a CT / PET imaging device 12 (and more specifically, a PET gantry 13) to acquire PET imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume (k) and a subsequent volume (k+1) that at least partially overlaps the volume (k) along the axial direction. At 104, at least one electronic processor 20 is programmed to generate an image of the volume (k) using an iterative image reconstruction process including operations 106 and 108. At 106, a local penalty function is calculated to suppress noise on the volume (k), including reducing the value of the local penalty function in the overlapping region. At 108, an updated image of the volume (k) is generated using the imaging data from the volume (k) and also using the local penalty function. Although not illustrated, it will be appreciated that the iterative image reconstruction 104 may account for radiation absorption by the object using an attenuation map generated from a CT image of the object acquired using the CT gantry 15 .

[0033] In one embodiment disclosed herein, the local penalty function depends on the local geometric sensitivity of the PET imaging device 12, and the value of the local compensation function is reduced in the overlap region by an amount that compensates for the loss of geometric sensitivity due to the axial truncation of the volume (k). An illustrative example is described next, in which the edge-preserving penalty function is a relative difference prior (RDP) and an ordered subset expectation maximization (OSEM) reconstruction is used. The updated image indexed n+1 for iteration n is appropriately written as:

[0034]

[0035] Among them, λ i is the estimated activity at the voxel indexed by i, and is the local penalty weighting factor, a jj′ is the system matrix value, s i,k is the geometric sensitivity at voxel i (preferably scaled by acquisition time if different bed positions have different acquisition times), y j is the data projection binning, and the parameter γ i > 0 is an edge preserving parameter and manipulates the RDP prior. The prior is the local image neighborhood N around voxel i i Usually, the larger the γ i Values ​​of γ produce greater edge preservation (i.e., reduce penalty). i = 0 will completely eliminate edge preservation, and RDP becomes a quadratic prior.

[0036] As disclosed herein, in the overlap region where the couch positions of volumes k and k+1 overlap (and in the overlap region where the couch positions of volumes k-1 and k overlap if the current couch position k overlaps at both ends), the value of the local penalty function is reduced. In some suitable embodiments, the local penalty function depends on the local geometric sensitivity s of the PET imaging device 12 to the voxel indexed by i in the image volume of the couch position of volume k. i,k The geometric sensitivity at voxel i can be considered as the number of lines of response (LOR) that pass through voxel i. In the part of volume k that does not overlap with adjacent volumes k-1 or k+1, the standard sensitivity can be used, and for the nth iteration, the value where γ0 is a constant. However, in the overlapping region, the penalty is reduced. This is compensated by increasing the value of the edge-preserving parameter γ in the overlapping region by an amount that compensates for the loss of geometric sensitivity due to the axial truncation of volume k. Therefore, a “joint” edge-preserving parameter is used It can be appropriately written as Among them, s i,k+1 is the geometric sensitivity of the (same) voxel i in the next neighboring volume k+1. Again, this adjustment increases the edge preservation parameter (i.e. ), which results in a reduction in the local RDP penalty in the overlapping region due to γ ​​being in the denominator of the local RDP penalty of equation (1). This reduction in the RDP penalty takes into account the additional data in volume k+1, which means that the noise is lower when weaving the two reconstructed volumes k and k+1 together in the overlapping region, so that a lower penalty can be employed in this overlapping region when reconstructing volume k to achieve the desired noise reduction, and thus features (edges) are better preserved due to the reduced penalty in the overlapping region.

[0037] In another embodiment, the local penalty function depends on the total coincidence count through the penalized voxel, and the value of the local compensation function is reduced in the overlap region by an amount that compensates for the total coincidence count through the penalized voxel in the successor volume k+1. The penalty can be formulated based on the actual local coincidence count statistics, rather than on the geometric sensitivity. This requires replacing the geometric sensitivity s in the above expression with the actual counts. In this case, (no overlap adjustment), where C i,k is the count of coincident events in the data set for bed position k whose LOR passes through voxel i. For time-of-flight (TOF) PET, this can be formulated as where index v traverses C of the dataset for bed position k whose LOR passes through voxel i i,k Matches the count, and w vis the time-of-flight probability of the count indexed by v originating from voxel i. For the non-TOF case, the joint edge-preserving parameter (i.e., providing overlap adjustment) is where, again similarly, term C i,k+1 is the count of coincident events in the data set for bed position k+1 whose LOR passes through the (same) voxel i. In this formulation, even more explicitly, the reduced penalty (which is again determined by the denominator term The increase in value provides) takes into account the additional data C in volume k+1 i,k+1 , which means that the noise is lower when weaving the two reconstructed volumes k and k+1 together in the overlapping region.

[0038]

[0039] Located in a dataset of volume k+1.

[0040] In the above example, the reconstruction process 106, 108 includes: reconstructing each volume (k) of imaging data separately from other volumes (k) of imaging data into a reconstructed couch position image, and then weaving the couch position images together in image space to generate a final image. In other embodiments, the reconstruction process 106, 108 may include combining the number of counts from all volumes (k) of imaging data into a total volume; and reconstructing the total volume into a reconstructed image. In the latter embodiment, the local penalty function depends on

[0041]

[0042] In either embodiment, if the acquisition times of consecutive bed positions are not all the same, the reduction in the value of the local penalty function in the overlapping region is also appropriately scaled based on the ratio of the corresponding acquisition times of volume (k) and the subsequent volume k+1.

[0043] refer to Figure 3, an illustrative embodiment of a respiratory gating imaging method 200 is graphically shown as a flow chart. This is a different application, but the disclosed method of adjusting the penalty in overlapping regions (now temporal, rather than spatial / axial) is similar. At 202, at least one electronic processor 20 is programmed to operate or control the PET imaging device 12 to acquire imaging data of a breathing patient. Concurrently with the PET imaging data acquisition, a respiratory gating signal is acquired by a respiratory monitor 201, which provides respiratory phase and time information. For example, the respiratory gating signal can be provided by the respiratory monitor 201, which includes a respiratory monitoring belt that measures the expansion and contraction of the chest during breathing; or the respiratory gating signal can be provided by the respiratory monitor 201, which includes a flow sensor that measures the air flow into and out of the mouth and / or nose; and so on. At 203, the PET data is binned by respiratory phase using the respiratory gating signal. By a method similar to Figure 2 , each respiratory phase bin is referred to by an index k, and two adjacent respiratory phase bins k and k+1 are adjacent in the respiratory sequence, i.e., in each respiratory interval, respiratory phase k+1 immediately follows respiratory phase k in time. It should be understood that binning by respiratory phase will cause PET data acquired at the same respiratory phase k in consecutive breaths to be combined to form a PET data set for respiratory phase bin k; similarly, PET data acquired at the same respiratory phase k+1 in consecutive breaths will be combined to form a PET data set for respiratory phase bin k+1; and so on. In addition, the bins overlap in time. At 204, at least one electronic processor 20 is programmed to generate an image of respiratory phase (k) using an iterative image reconstruction process that operates on the acquired PET imaging data accumulated in respiratory phase bin (k) over multiple breaths. Iterative reconstruction 204 includes operations 206, 208, and 210. At 206, the imaging data of each bin is adjusted for respiratory motion detected in the gate phase. This requires shifting the counts of the LOR endpoints according to the derived motion vectors. For each voxel i, there is an associated motion vector Δr i , indicating the voxel i from its reference position (e.g., its position at bin k = 0 can be arbitrarily specified so that for all i, Δr i,k＝0 =0). At 208, a local penalty function is calculated to suppress noise in respiratory phase bin (k), including reducing the value of the local penalty function in the region of temporal overlap with respiratory phase bin k+1. At 210, an updated image of respiratory phase bin (k) is generated using imaging data from respiratory phase bin (k) and also using the local penalty function.

[0044] In some embodiments, the adjustment at operation 206 is performed according to the following formula:

[0045]

[0046] where γ is the edge preservation parameter, i is the voxel, λ is the grayscale value of the voxel, s is the geometric sensitivity across the voxel, and Δr is the phase of the two gates φ k and φ k+1 The motion vector between .

[0047] Figure 4 An example of a reconstructed image using a uniform phantom is shown, which has six 6 mm diameter spheres in a hexagonal pattern located in the center of the overlapping area. The figure marked (a) is an image reconstructed by standard penalized reconstruction after weaving. The figure marked (b) is an image reconstructed by joint penalized reconstruction of imaging method 100. A side view of the model setup is shown in the figure marked (c). A cylindrical model 50 is shown, which has multiple spheres 52 on a first bed position 54 and a second bed position 56.

[0048] The present disclosure has been described with reference to preferred embodiments. Modifications and changes may occur to others after reading and understanding the preceding detailed description. The present disclosure is intended to be interpreted as including all such modifications and changes as long as they fall within the scope of the appended claims or their equivalents.

Claims

1. A non-transitory computer readable medium storing instructions capable of being read and executed by a workstation (18), the workstation comprising at least one electronic processor (20) to perform an imaging method (100), the method comprising: receiving imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume k and a subsequent volume k+1 at least partially overlapping the volume k along the axial direction; and An image of the volume k is generated using an iterative image reconstruction process, wherein an iteration of the iterative image reconstruction process comprises: Calculating a local penalty function for suppressing noise over the volume k, comprising reducing a value of the local penalty function in an overlap region, the local penalty function depending on a total coincidence count through penalized voxels, and the value of the local penalty function being reduced in the overlap region by an amount compensating for the total coincidence count through the voxels penalized in the successor volume k+1; An updated image of the volume k is generated using imaging data from the volume k and also using the local penalty function.

2. The non-transitory computer readable medium of claim 1, wherein: The local penalty function depends on the local geometric sensitivity of the PET imaging device (12), and the value of the local penalty function is reduced in the overlap region by an amount that compensates for the loss of geometric sensitivity due to the axial truncation of the volume k.

3. The non-transitory computer readable medium of claim 2, wherein: The local penalty function depends on where γ is the edge-preserving parameter of the local penalty function, i indexes a voxel, is the value of the voxel i in the iterative reconstruction of the volume k, specifically the value of the voxel i in the updated image n of the iterative reconstruction n, s i,k is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k, and s i,k+1 is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k+1.

4. The non-transitory computer readable medium of claim 1, wherein: The penalty function depends on where γ is the edge-preserving parameter of the local penalty function, i indexes a voxel, is the value of the voxel i in the iterative reconstruction of the volume k, specifically the value of the voxel i in the updated image n of the iterative reconstruction n; C i,k is the count of coincident events in the dataset for the bed position in volume k where the LOR passes through that voxel i.

5. The non-transitory computer readable medium of claim 2, wherein: The PET imaging device (12) is a time-of-flight imaging device, and The penalty function depends on where γ is the edge-preserving parameter of the local penalty function, i indexes a voxel, is the value of the voxel i in the iterative reconstruction of the volume k, specifically the value of the voxel i in the updated image n of the iterative reconstruction n; C i,k is the count of coincident events in the dataset for bed positions k whose LOR passes through voxel i; and index v iterates through the C count of coincident events in the dataset for bed positions k whose LOR passes through voxel i. i,k The number of matches is , and w v is the time-of-flight probability that the count indexed by v originates from the voxel i.

6. The non-transitory computer readable medium of any one of claims 1 to 5, wherein: The reconstruction includes: The volume k of the imaging data is reconstructed separately from other volumes k of the imaging data as a reconstructed image.

7. The non-transitory computer readable medium of any one of claims 1 to 5, wherein: The reconstruction includes: combining the number of counts from the volume k in the imaging data into a total volume; and The total volume is reconstructed into a reconstructed image.

8. The non-transitory computer readable medium of claim 2, wherein: The local penalty function depends on where γ is the edge-preserving parameter of the local penalty function, i indexes a voxel, is the value of the voxel i in the iterative reconstruction of the volume k, specifically the value of the voxel i in the updated image n of the iterative reconstruction n, s i,k is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k, and s i,k+1 is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k+1.

9. The non-transitory computer readable medium of any one of claims 1-5 and 8, wherein: The reduction in the value of the local penalty function in the overlap region is scaled based on a ratio of respective acquisition times for the volume k and the successor volume k+1.

10. An imaging system (10), comprising: Positron emission tomography (PET) imaging equipment (12); as well as At least one electronic processor (20) programmed to: receiving imaging data on a frame-by-frame basis for frames along an axial direction, wherein adjacent frames overlap along the axial direction, wherein the frames include at least a volume k and a subsequent volume k+1 at least partially overlapping the volume k along the axial direction; and An image of the volume k is generated using an iterative image reconstruction process, wherein an iteration of the iterative image reconstruction process comprises: Calculating a local penalty function for suppressing noise over the volume k, comprising reducing a value of the local penalty function in an overlap region, the local penalty function depending on a total coincidence count through penalized voxels, and the value of the local penalty function being reduced in the overlap region by an amount compensating for the total coincidence count through the voxels penalized in the successor volume k+1; generating an updated image of the volume k using imaging data from the volume k and also using the local penalty function; The reconstruction includes: combining the number of counts from the volume k in the imaging data into a total volume; and The total volume is reconstructed into a reconstructed image.

11. An imaging method (100) performed on imaging data on a frame-by-frame basis for frames along an axial direction, wherein: Adjacent frames overlap along the axial direction, wherein the frames include at least a volume k and a subsequent volume k+1 at least partially overlapping the volume k along the axial direction, the method comprising: An image of the volume k is generated using an iterative image reconstruction process, wherein an iteration of the iterative image reconstruction process comprises: Calculating a local penalty function for suppressing noise on the volume k, comprising reducing a value of the local penalty function in an overlap region, the local penalty function depending on a local geometric sensitivity of a PET imaging device, and the value of the local penalty function being reduced in the overlap region by an amount compensating for a loss in geometric sensitivity due to an axial truncation of the volume k; generating an updated image of the volume k using imaging data from the volume k and also using the local penalty function; The reconstruction includes: The volume k of the imaging data is reconstructed separately from other volumes k of the imaging data as a reconstructed image.

12. The imaging method (100) according to claim 11, wherein: The local penalty function depends on where γ is the edge-preserving parameter of the local penalty function, i indexes a voxel, is the value of the voxel i in the iterative reconstruction of the volume k, specifically the value of the voxel i in the updated image n of the iterative reconstruction n, s i,k is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k, and s i,k+1 is the geometric sensitivity of the PET imaging device (12) to the voxel i when acquiring data over the volume k+1.

13. The imaging method (100) according to any one of claims 11 and 12, wherein: The reduction in the value of the local penalty function in the overlap region is scaled based on a ratio of respective acquisition times for the volume k and the successor volume k+1.

14. An imaging method (200) performed on imaging data on a frame-by-frame basis for frames along an axial direction, wherein: Adjacent frames overlap along the axial direction, wherein the frames include at least a volume k and a subsequent volume k+1 at least partially overlapping the volume k along the axial direction, the method comprising: An image of the volume k is generated using an iterative image reconstruction process, wherein an iteration of the iterative image reconstruction process comprises: conditioning the imaging data with respiratory gating using two or more gate phases and motion detected in the gate phases; calculating a local penalty function for suppressing noise over the volume k, including reducing a value of the local penalty function in an overlapping region; An updated image of the volume k is generated using imaging data from the volume k and also using the local penalty function.

15. The method (200) of claim 14, wherein: The adjustment is performed according to the following equation: where γ is the edge preservation parameter, i is a voxel, f is the grayscale value of the voxel, s is the geometric sensitivity across the voxel, and Δr is the phase of the two gates φ k With φ k+1 The motion vector between .

Images