Data-Driven Mitigation of SPECT Motion-Induced Artifacts White Paper

White paper Data-driven mitigation of SPECT motion-induced artifacts Maximilian P. Reymann, MSc Brandon C. Jones, PhD Francesc Massanes, PhD A. Hans Vija, PhD siemens-healthineers.com/symbia-prospecta SIEMENS Healthineers Table of contents Overview 3 Key motivation 4 Characteristics of gamma cameras with multi-head collimation 6 Intra-view motion correction with rMC Pro 7 Respiratory motion correction theory 7 rMC Pro validation 8 Inter-view motion correction with MC Pro 9 Data model consistency theory for MC Pro 9 MC Pro inter-view correction with xSPECT Bone™ 10 MC Pro validation 11 rMC Pro and MC Pro: clinical validation 12 rMC Pro and MC Pro: clinical workflow and tips 14 Overview of motion correction workflow on Symbia Pro.specta™ 14 Recommendations for rMC Pro 15 Recommendations for MC Pro 15 Discussion and summary 16 About the lead author 17 References 18 2 White paper · Data-driven mitigation of SPECT motion-induced artifacts Overview Symbia Pro.specta SPECT/CT introduced two motion-mitigation technologies that can improve iterative reconstruction by detecting and correcting for bulk motions between views (also known as MC Pro) and for intra-view respiratory motion (also known as rMC Pro). These novel methods do not require any additional hardware and are purely algo- rithmic, data-driven improvements to ensure maximum patient comfort and seamless integration into existing clinical workflows. Furthermore, MC Pro and rMC Pro can be enabled with the click of a button in the Exam Designer page or in the Exam Workflow on the Symbia Pro.specta™ user interface (see Figure 1) and do not require the user to perform manual corrections. This white paper reviews key pertinent information related to MC Pro and rMC Pro, including key features and underlying theory, validation in phantom and clinical experiments, and presents guidelines for use and some potential pitfalls. Back to Exam Designer Scan Protocols 1. Select Protocols 2. Modify Protocols 3. Confirm Changes Protocol Scan Favorites General Scan Gantry Settings Dose Timing Config Physio Scan Contrast Recon Favorites General Recon/ Recon&GO Image Impression Recon Box Physio Recon Auto Tasking Inline Options Find/Replace can/Recon General Recon Scan/Recon Basic Recon Recon Scatter Mode Method Correction Attenuation Correction Intra-Recon Respiratory Match CT Motion C ... Motion C ... Slices Image Order Slice Thickness Incremen No. of [mm] Comment [mm] Images Cardiac SPECT-CT [factory] (Adult) Tomo Tomo Stress Tc-99m Tomo Projections Cardiac Flash 30+ (3-1) CT-AC 3.00 Qr40 53 On Feet to Head 3.00 3.00 AC Recon Gated Tomo Stress Tc-99m MC Flash 30+ On (3-1) CT-AC 3.00 Qr40 53 On On Feet to Head 3.00 3.00 NoAC Recon Tomo Stress Te.99m MC Cardiac Flash 30+ OT None Feet to Head 3.00 3.00 NoAC Recon Gated Tomo Stress Tc-99m MC Cardiac Flash 30+ None Feet to Head 3.00 3.00 Topogram Topogram 0.70 Tr20 cor CT-AC CT-AC 3.00 Qr40 53 Feet to Head 3.00 3.00 129 Select Protocols Validate Scan Mode Confirm Changes Figure 1: Exam Designer page in VA20, demonstrating the MC Pro and rMC Pro workflow. The red box highlights configurable options for MC Pro and for rMC Pro. These data-driven motion corrections are enabled with simple recon parameter selections by the user. White paper · Data-driven mitigation of SPECT motion-induced artifacts 3 Key motivation An image is the multidimensional aggregation of camera sensor data when it has been processed to meet a specific task. Single-photon emission computed tomography (SPECT) images are created by “processing” sequentially acquired projection view data with tomographic image reconstruction algorithms to produce a spatial map, ie, image, of the radioisotope distribution. While data is what the SPECT instrument acquires, the data model is the “data” generated by the modeling of an actual instrument, in essence the image reconstruction problem which simulates the acquisition. Images could, themselves, become data too, if they are later used for a subsequent task. While data and images are sometimes used synonymously, one must be careful to differentiate between the two when discussing image artifacts. In the following text, the term “data” for SPECT specifically refers to the data space or 2D projection “images” as measured by the gamma camera at a specific angle (also known as one “view”), whereas the term “data model” refers to the data processing and image reconstruction methodology. An image artifact is a component of the image in the context of a specific clinical task, which does not match expectations or reality. Image artifacts can arise either from (1) an inconsistency between the measured data and the data model or from (2) data artifacts, which are typically a corruption of the raw data such that the information contained in the data (using the expanded definition of data above) is incomplete or wrong. However, the latter is far less common currently, while the former inconsisten- cies between data and the data model are the primary cause of image artifacts. This specific distinction is important, as the current SPECT image formation methodology resulting in a stack of 2D projection images is inherently tomographically consistent, if the radiotracer uptake does not change during the time of the acquisition. Tomographic consistency necessitates consistency within one view (ie, “intra-view”) and between many views (ie, “inter-view”) and is a fundamental assumption of the SPECT data modeling. Motion, therefore, violates this assumption and creates a source of inconsistency between the data and data model, which can manifest as a “smear” in one or more views (intra-view), or as a relative shift in the activity region between views (inter-view). Relative shifts can be observed because SPECT image formation obtains projection data from viewing angles in a temporal sequence, unlike positron emission tomography (PET), which can simultaneously acquire all viewing angles for tomography. There are many other potential sources of inconsistency between SPECT data and the data model, eg, lack of attenuation correction, or mis-registration between the CT and the uncorrected nuclear image. 4 White paper · Data-driven mitigation of SPECT motion-induced artifacts Shifting projection views in the axial and trans-axial directions to improve data consis- tency with the data model has been a common practice in SPECT imaging for decades.1 Classical approaches relied on the assessment of sinograms and linograms, either visual or semi-automatic, to determine the shift amount so that motion artifacts are mitigated. Once the artifacts have been corrected, reconstruction starts, either as single back projection, as in filtered back projection (FBP), or as an iterative reconstruction (IR). Unlike filtered back projection, where only a single, global filtering operation is possible—in iterative reconstruction, the estimation of the reconstructed image based on the acquired data is sequentially updated. This effectively improves the resolution as the number of iterations increases, if the data model is a good approximation of the data; however, in the case of a data versus data model mismatch, the errors or artifacts also become enhanced with increasing number of updates. This indicates a trade off is needed between image resolution and artifacts. Depending on the clinical task, the image artifact might be relevant and need to be suppressed, or the image resolution might not be the driving criteria. The alternative option is to alter the data model itself to improve agreement with the measured data. In the case of SPECT motion correction, a complete correction of all motion would imply a local non-rigid model with rotation and translation in all three dimensions, which is not possible because SPECT acquires a limited set of projection views in a temporal sequence, and the collimated image formation physics poorly differentiates the impact of certain types of motion, eg, motion orthogonal to the detector plane. However, SPECT imaging is nonetheless sensitive to specific types of motion, eg, motion parallel to the detector plane, and these are well-suited for state-of-the-art motion correction schema, as detailed below. The key motivation for developing a new motion correction approach was to create one that is compatible with multimodal quantitative SPECT requiring iterative reconstruc- tion. Only iterative reconstruction enables a rigorous implementation of the image for- mation physics, which is a prerequisite for absolute quantitative SPECT. Quantitative imaging requires complete traceability of measurement values and errors for all steps from acquisition to reconstruction, which precludes any human intervention for man- ual/semi-automatic corrections. Furthermore, even with non-quantitative iterative reconstruction, conventional motion correction approaches involving manual/semi-auto- matic shifts become inherently impractical due to poor reproducibility and the excessive time required to make a correction, attempt a reconstruction, evaluate the result, and potentially revert to shift views and repeat the process.2 There is a clear need for data-driven and iterative reconstruction-compatible technolo- gies which mitigate motion artifacts by improving consistency between the SPECT data and data model and, by extension, the image quality and resolution.3-6 The data-driven motion correction methods presented here allow the reconstruction engine to autono- mously detect motion characteristics and apply mitigations in the form of axial and trans-axial data shifts to improve the iterative reconstruction. White paper · Data-driven mitigation of SPECT motion-induced artifacts 5 Characteristics of gamma cameras with multi-head collimation While the following applies in principle to all parallel or focusing collimated image formation, where the needed views for tomography are acquired in a temporal sequence, here we focus on multi-head cameras like Symbia Pro.specta SPECT/CT. The dwell time of a view refers to the period during which data is collected for a specific view. During each dwell time, a set of views is acquired as vi [v1 = vj vNd ], where , ... , ... , Nd is the number of detectors (Nd 2 for Symbia Pro.specta), and an acquisition A = consists of multiple views Nv such that A = [v1 vi vNv ]. Any motion that is con- , ... , ... , tained in the acquisition causes an inconsistency between assumed data distribution and actual data distribution. As discussed above, SPECT systems with large field of view (FOV) sodium iodide (NaI) detectors are sensitive to axial shifts (shift in y) and trans- axial shifts (shifts in x) of the detector coordinate system (see Figure 2). However, motion that is orthogonal to the detector surface, ie, moving towards/away from the detector, causes a ‘soft’ change of resolution in the data that is not readily detected or corrected using parallel hole collimators. If astigmatic collimators are used, motions in the z direction could produce more prominent changes in count distribution depending on the distance and orientation and design of the collimator, but this adds additional degrees of freedom and further complicates the reconstruction with SPECT-only data. Fundamentally, data-driven motion correction methods can only correct for motions that are visible in the data. Therefore, we restict motion identification and correspond- ing corrections to those SPECT systems that are especially sensitive to axial and trans- axial translations. SIEMENS ; Healthineers . Figure 2: The left illustration shows an overview of the z system with the patient bed. y x The right panel shows the local z detector coordinate system in ....... y x Symbia Pro.specta. When the patient is positioned on the . . . ........ bed, the x-axis corresponds to the left-right orientation, while the y-axis corresponds to the head-toe direction. 6 White paper · Data-driven mitigation of SPECT motion-induced artifacts Intra-view motion correction with rMC Pro Respiratory motion correction theory Motion that occurs within the dwell time of one view causes blurring or smearing within the view. While there are myriad causes for intra-view motion, eg, respiratory, cardiac, peristalsis, coughing, tremors, etc, respiration is most commonly the dominant source of intra-view motion.1 Respiratory motion correction can be achieved through gating based on similar respiratory phases, which necessitates an estimation of the respiratory motion. Although several external devices exist to record patient respiratory motion, clinical acceptance of these devices has been low, as it represents additional burden for the patient and the clinical staff. Therefore, rMC Pro on Symbia Pro.specta uses a data-driven gating (DDG) method to extract a respiratory signal from the data directly, thus promoting patient comfort and easy clinical workflows. Since respiratory motion has a predominantly superior-inferior motion component, it is mainly visible as y-axis (axial) shifts on the detector, assuming the patient is oriented along the long axis of the scanner. Depending on the position and orientation of the detector gantries, the anterior-posterior expansion of the patient during respira- tion may also be visible as x-axis (trans-axial) shifts on the detector. However, since detection of the anterior-posterior expansion is unreliable and noisy, this component is ignored, and presumably, respiratory motion predominantly occurs on the y-axis of the detector. The rMC Pro methodology can be applied to cardiac acquisitions acquired with IQ•SPECT™ technology and SMARTZOOM™ collimators, as well as low-energy, high-res- olution (LEHR) collimators from either continuous or step-and-shoot rotations, provided sufficient counts are maintained. The one-dimensional respiratory waveform used for gating is estimated through a non-linear manifold learning method that extracts a respi- ratory surrogate signal from finely sampled frames of the nuclear medicine raw data using the Laplacian Eigenmaps (LE) algorithm.7 First, the nuclear medicine raw data for the given view is framed to 200 ms bins containing the counts of all energy windows and both detectors, arranged next to each other. To reduce the impact of noise on the LE method, this data is then temporally and spatially smoothed. Next, for each view, a respiratory surrogate signal r(t) is estimated based on the lower-dimensional repre- sentation obtained from the LE method of the smoothed data of both detectors. As this estimate tends to be still very noisy, a post-processing pipeline is applied to convert the extracted respiratory surrogate signal into a normalized, smoothed, and correctly oriented respiratory surrogate signal that can be used for gating within this view. The respiratory gate for each time frame in a view is then obtained via amplitude binning with Ng = 6 gates. White paper · Data-driven mitigation of SPECT motion-induced artifacts 7 Once the view has been split into respiratory gates, independent axial shifts are computed from the nuclear medicine raw data to move each gate to the overall view’s reference (see Figure 3). Specifically, an intra-view shift vector matrix Sintra ∈ ℝNv x Ng is calculated that represents how much the y-median of a specific respiratory gate deviates from the overall y-median of its respective view via the equation: Sintra = Median(yv,g) – Median(yv v,g ). where, yv,g refers to the vector y-coordinates of all measured counts in gate g and view v and yv to the vector of y-coordinates over all gates in view v. Using the y-median of the entire view improves the statistical power by using all available counts and creates a more robust solution that is less sensitive to outliers than if the reference were restricted to a single physiological respiratory phase such as end-exhale. As noted, rMC Pro cor- rections are applied to the nuclear medicine raw data space and all subsequent steps including MC Pro and image reconstruction use this corrected nuclear medicine raw data. Additionally, since this respiratory mitigation method is applied within each view, it does not by itself account for respiratory effects seen across views. For example, in the case of varying respiratory phases between successive views, especially when dwell time matches the respiration period, there can be oscillations or “jitters,” which shift the FOV between views. However, although this type of motion is still attributable to “respi- ration,” in this implementation, this type of motion is not addressed by rMC Pro but would, instead, be mitigated by using inter-view correction MC Pro, as detailed below. Figure 3: Illustration demon- strates the rMC Pro methodol- ogy for intra-view respiratory Respiratory gates motion corrections. This is (uncorrected) exaggerated for visualization purposes. In the top row, the y-axis axial correction vectors (teal arrows) point from each respiratory gate y-median (orange circles) to the entire Respiratory gates view’s y-median (black axis (corrected) line). The middle row illustrates the corrected respiratory gate y-medians (green circles) after aligning to the view median (black line). The bottom row Respiratory gate duration 1 2 3 4 5 6 depicts approximate temporal contribution for each gate. rMC Pro validation The data-driven respiratory signal extraction methodology based on LE algorithm has been detailed and validated previously.7 In this study, 67 patient scans from 3 different radiotracer/scan protocols (27 99mTc-Sestamibi rest/stress myocardial perfusion, 26 99mTc-MAA shunt diagnostic, and 14 99mTc-DTPA-MAA lung perfusion scans) were performed along with a pressure sensor belt (Anzai AZ-733V) affixed to the patient, which continuously measured respiratory signal for the duration of the scan. The pressure sensor respiratory signal served as a ground truth validation for the data- driven estimated waveform. Notably, the data-driven respiratory signal achieved a 8 White paper · Data-driven mitigation of SPECT motion-induced artifacts high significant Pearson correlation coefficient with the ground truth with an average ± standard deviation across scans of 0.81 ± 0.17 and a median of 0.89. Additionally, it demonstrated that the quality of the respiratory signal prediction was associated to the count density of the scans, with high-count scans producing high-quality res- piratory signal predictions. Furthermore, even in some cases with irregular respiratory signals arising from cardiac diseases, the data-driven method still produced reasonable results and reliably captured the overall structure of the respiratory signal.7 Inter-view motion correction with MC Pro Data model consistency theory for MC Pro To correct motion that manifests as apparent shifts between views, the Improved Tomographic Consistency (ITC) method4 was developed, available as MC Pro on Symbia Pro.specta SPECT/CT. As the name implies, rather than correcting for motion in the data, this method aims to automatically improve consistency between the SPECT data and data model by adapting the data model to match the measured data. Conceptually, this is performed by computing a preliminary “anchor” image recon- struction with only a small number of updates and then comparing each measured view of the data with the corresponding forward projection of the “anchor” image. By employing rigid image registration to compute axial and trans-axial shifts between the measured view (data) and the forward projected anchor image (data model) for each view, the data-to-data model consistency is improved and the effect of motion between views is mitigated (see Figure 4). Figure 4: Illustration demon- strates MC Pro methodology for inter-view irregular bulk motion corrections. This figure is exag- SPECT or SPECT/CT xSPECT Bone gerated to illustrate the meth- odology. The attenation- and scatter-corrected SPECT recon- data model data model struction (on the left), or the Projected Projected zone map derived for xSPECT Bone (on the right), serves as Data Data the “anchor image.” For each view, the anchor image is for- SPECT data Zoned data ward projected (teal arrows) “Anchor image” “Anchor image” into the view data model and then 2D rigid registration is per- formed to determine optimal axial and trans-axial shift vec- tors (orange arrows) to align the data and correct for motion. Forward projection of data 2D registration between Data courtesy of model for each view data model and data Universitätsspital Basel, Basel, Switzerland. Data on file. White paper · Data-driven mitigation of SPECT motion-induced artifacts 9 The method assumes the primary source of inconsistency results from the misregistra- tion, which is required to determine the optimal axial and trans-axial shifts to improve data-to-data model agreement. However, in practice, this is not the case, and it is more common that attenuation and scattering are the dominating sources of change between views. In this case, estimating shift vectors would be futile. Therefore, MC Pro first creates an attenuation- and scatter-corrected anchor image that is stopped early with only a quarter of the required iterations. In the case that a CT scan is present, the CT scan is used for attenuation correction; however, if there is no CT scan present, a separate data-driven attenuation map is estimated purely for the purpose of mitigating attenuation effects in the MC Pro anchor reconstruction, and then the data-driven attenuation map is later discarded. Once the attenuation- and scatter-corrected MC Pro anchor image is generated, the algorithm will iterate through each view and compare the forward projected view (data model) to the measured projection (data). The inter-view shifts Sinter ∈ ℝ2xNdxNv are determined by employing rigid registration to identify the axial and trans-axial shifts between estimated data and measured data, for the entire FOV of each detector of a view independently. This can then be repeated for several iterations, refining the shift vectors during each iteration, resulting in all acquired data registered to the same data model. It is important to note that a single axial and trans-axial shift is computed for all data in a detector view vj i (both detectors are handled separately) on the assumption that there was some uniform global bulk motion between views. However, in the case where a view contains multiple structures characterized by differ- ent motion sources (consider a static spine, but a moving ribcage), corrections would be computed based on the dominant motion source and those would be applied to the entire view. So, if the moving ribcage is identified as the primary source, then the cor- rections would also shift the static spine. Conversely, if the motion is only present in a small structure within the view, MC Pro might calculate that minimal shifts are needed for the whole view. MC Pro and rMC Pro therefore require careful consideration of the clinical application in question and the acquired data quality to determine if it is right to use. Effectively, MC Pro should only be used if motion is visible in the target organ/area of interest. Moreover, the registration can be obfuscated through local changes in the visible tracer distribution that are not or only partially visible in the data model. MC Pro inter-view correction with xSPECT Bone For xSPECT Bone, the CT scan is the frame-of-reference from which a zone map is derived.8 This zone map then serves as the dedicated data model “anchor image” in place of the SPECT reconstruction. For each view, the zone map anchor image is forward projected to data space (data model) and then compared to the measured view (data) to compute the shift vectors via 2D registration.4 10 White paper · Data-driven mitigation of SPECT motion-induced artifacts MC Pro validation The methodology used in MC Pro has been detailed and validated previously.9,10 The MC Pro method was first demonstrated in a study of 21 99mTc diphosphonate bone scans with no motion.9 Synthetic perturbations were applied to the acquired data by randomly shifting projection views up to 10 mm and then reconstructions were performed with a prototype version of MC Pro (MC Pro group) or with no corrections (high-motion group) applied. Data was assessed qualitatively with visual inspection and quantitatively by computing the local modified cumulative density function (MCDF) score as a measure of the tomographic consistency.11 Reconstructions with a prototype MC Pro demonstrated clear visual improvement of motion artifacts. The MC Pro group had significantly better MCDF scores than the high-motion group and were similar to the original motion-free group (MC Pro group: 5.7 ± 3.7%, high motion: 14.6 ± 4.9%, low motion: 5.4 ± 3.5%; values reported as average ± standard deviation across scans). Statistical comparisons demonstrated that motion was corrected in 19 cases and was mitigated in the remaining 2 cases. Notably, applying the MC Pro methodology to cases with no synthetic shifts did not induce artifacts in the data.9 The efficacy of the MC Pro method was further demonstrated in a large study of 165 clinical SPECT scans drawn from two cohorts.10 The first cohort comprised 76 99mTc bone diphosphonate SPECT subjects with clear visible motion artifacts present in the data (motion group), and the second cohort comprised 89 scans with no visible motion from either 99mTc bone SPECT scans or 99mTc lung perfusion SPECT scans (coached group). Images were reconstructed with xSPECT™ quantitative reconstruction with attenuation and scatter correction. Visual assessment indicated that application of a prototype MC Pro version mitigated motion artifacts in scans with known motion and had no notice- able effect on scans without motion. Remarkably, all reconstructions exhibited improved numerical MCDF scores after motion correction, which suggests high tomographic con- sistency (see Figure 5). Notable quantitative improvements in image quality were seen in both cohorts with and without apparent motion present in the initial scans, with larger improvements seen in cases with greater motion. Importantly, the scans with no motion were not negatively impacted by the application of the MC Pro technique.10 White paper · Data-driven mitigation of SPECT motion-induced artifacts 11 Figure 5: Example of prototype MC Pro methodology applied to two clinical cohorts, one with motion (motion group) and one 0.30 xSPECT-CG with low motion (“coached” xSPECT-CG + MC Pro cohort), as detailed in10. Two reconstructions, one with MC Pro on and one without motion correction, were generated for each scan using the quantita- tive xSPECT Quant™ method 0.20 with attenuation and scatter correction. Image quality Note: This is a assessed with a score derived prototype version from the MCDF metric demon- of MC Pro strated substantial reduction MCDF score 0.10 in motion from the high motion group with the application of MC Pro, whereas the low motion group showed no significant differences with the application of MC Pro.11,12 This analysis indi- 0 HIH HIH cates that MC Pro will not mark- Motion group Coached group edly impact the reconstruction if no motion is detected. rMC Pro and MC Pro clinical validation The combined clinical benefit of prototype versions of rMC Pro intra-view corrections and MC Pro inter-view corrections were shown for SPECT myocardial perfusion imaging (MPI).2 As changes in the polar map might be difficult to visually assess in vivo, we applied the shift vectors obtained from the patient data to a still phantom, thus demon- strating the relative difference they can cause in the reconstructions (see Figure 6). Furthermore, this study also compared a cardiac-focused prototype including MC Pro and rMC Pro to conventional SPECT (Flash 3D) in 56 patients with suspected coronary heart disease. The prototype utilized acquisitions with nuclear medicine raw data recording, continuous rotation, retrospective ECG gating, CT-based attenuation correc- tion (AC), energy window-based scatter correction (SC), and prototype intra-view rMC Pro and inter-view MC Pro motion corrections. Results showed the prototype produced more homogenous myocardial polar distributions, higher image contrast, and better utilization of ECG beats (see Figure 7). Images were rated by physicians as having superior (44%) or comparable (56%) image quality compared to the conventional approach. The addition of motion correction was shown to have superior image contrast in identifying myocardial perfusion defects (0.48 for the prototype vs 0.56 for conventional, p=0.0004, here lower values indicate greater contrast). Overall, this data indicates the clinical benefit of rMC Pro and MC Pro data-driven motion correction methods in quantitative SPECT imaging. 12 White paper · Data-driven mitigation of SPECT motion-induced artifacts Application of inter and intra-view motion to no-defect phantom Degree of respiratory motion in each respiratory gate Inter-view motion to x- and y-direction in a patient HTH HHO oo LA . OH Ho Number Motion (mm) . HIH -8 -10 oo OHH Respiratory gate number Motion (mm) Motion (mm) Figure 6: Application of intra- A. Still phantom B. Respiratory motion C. Respiratory motion 2x D. Motion to x-y directions (rMC Pro) and inter-view (MC Apply Pro) motion correction to a motion no-defect phantom. Repro- vectors duced from Nakajima et al.2 Licensed under a Creative Commons Attribution 4.0 International License Patient-specific inter-view notion to x- and y-directions (http://creativecowmmons.org/ licenses/by/4.0/) (A) Small inferior infarction with OMI (B) Lateral infarction with left ventricular dilation Flash 3D Figure 7: Example of two patients scanned using two modes: the current clinical standard of care and the cardiac-focused prototype. xSC (No ACSC) Left panel: Small heart with inferior old myocardial infarc- tion. The small inferior wall infarct visualization is improved in attenuation/scatter-corrected xSC (ACSC) image using the prototype. C. Right panel: Lateral infarction with left ventricular dilation. The site of infarction is better defined with the prototype with inter- and intra-view corrections. xSC (ACSC + inter and intra-view correction) Reproduced from Nakajima et al.2 Licensed under a Creative Commons Attribution 4.0 International License (http://creativecommons.org/ licenses/by/4.0/) White paper · Data-driven mitigation of SPECT motion-induced artifacts 13 rMC Pro and MC Pro: clinical workflow and tips Overview of motion correction workflow on Symbia Pro.specta Figure 8 shows the workflow of rMC Pro and MC Pro on Symbia Pro.specta SPECT/CT. Importantly, when rMC Pro is enabled, intra-view respiratory corrections are applied first in the nuclear medicine raw data space, followed by MC Pro, where corrections are applied on a view-by-view basis on framed data. Both motion correction technologies require sufficient counts in the acquired images to work. If the acquired data has low counts, there will not be enough statistical power to discern inconsistencies and estimate corrections, and application of these methods could further degrade the images. Furthermore, these methods assume that there is rigid motion present, and that it exists in a direction sensitive to SPECT imaging (axial for rMC Pro, axial and trans-axial for MC Pro). In the case of data inconsistencies arising from other sources or from complex motion (eg, rotations, out-of-plane translations, non-rigid motion), these will not be corrected for, and using these methods could intro- duce additional sources of bias. Additionally, the system does not know which specific parts or subregions of the image are of interest for the current acquisition, so correc- tions are applied to the entire view FOV. Acquired data Acquired data Corrected data Reconstruction Apply respiratory Apply inter-view T motion correction? motion correction? rMC Pro Preliminary Get shift reconstruction Iterate vectors Figure 8: Overview of motion correction workflow on Symbia Register estimated data Pro.specta SPECT/CT. rMC Pro (data model) to for respiratory corrections and measured data MC Pro for inter-view motion correction are enabled with the click of a button on the user interface. 14 White paper · Data-driven mitigation of SPECT motion-induced artifacts Recommendations for rMC Pro Respiratory motion within projection views can be mitigated with rMC Pro, provided there is sufficient counts within the acquisition to the extent the data has enough statistical power, ie, counts. rMC Pro can be used as default for myocardial perfusion imaging if the heart is the dominant source of activity within the FOV for each frame. The most common confounding situation for rMC Pro methodology is when the heart is not the strongest source of activity within the FOV. In this case, the algorithm could pick up on the motion of the hotter organ and apply corrections based on this. Therefore, users should carefully consider confounding sources of motion within the FOV for all view angles. Additionally, certain site-, patient-, and scan-preparation protocols can help mitigate confusing activity outside of the organ of interest. Finally, it is important to reiterate that, while rMC Pro is designed for “respiratory correction,” this is performed within views, and respiratory motion between views will instead be handled by MC Pro. Recommendations for MC Pro Irregular bulk motion between projection views can be mitigated with MC Pro, provided there is sufficient counts within the acquisition. MC Pro should be used only when there is motion present within the clinical organ of interest. MC Pro motion corrections apply a single axial and trans-axial translation to each projection FOV used in the recon- struction that does not account for separate regional or tissue compartmental motion. Therefore, in the case that there are multiple moving tissue compartments or a moving compartment and a static one, applying MC Pro could improve the dominant moving compartment while blurring the other. However, it is possible in some cases to mini- mize confounding effects from other tissues by restricting the reconstruction FOV to focus on the clinical target organ of interest while cropping out the source of confusing activity. Additionally, MC Pro methodology can also be confused by changes in radiotracer distribution over the course of the scan and works best when the biodistribution is static. MC Pro therefore requires careful consideration of the clinical application in question as well as the acquired data quality to determine if it is right to use. White paper · Data-driven mitigation of SPECT motion-induced artifacts 15 Table 1: Recommendations for motion correction methods rMC Pro MC Pro on Symbia Pro.specta. Summary 1. Can be turned on to 1. Can be turned on to correct correct for intra-view for jitter between views respiratory motion 2. Corrections are determined 2. Corrections are determined using registration of data to by comparing median loca- “anchor image” data model tion within each respiratory for each detector and view gate to the entire view 3. Corrections are applied to 3. Corrections are applied all counts within a view to all counts within a view 4. Shifts entire view uniformly, in nuclear medicine raw can therefore introduce data space inaccuracies to regions that originally had no motion What is Location of respiratory Location of each corrected gates within a view projection view Type of Axial direction Axial and trans-axial direction corrections When to use Ideal when the clinical organ of Should only be applied when interest is the dominant source clinical organ/area of interest of activity in each frame has motion Image quality Works better with Works better with higher counts higher counts Discussion and summary The use of intra-view and inter-view motion corrections in SPECT helps recover resolution and reduce artifacts from mismatched data and data model. Although the inter-view approach is technically not motion “correction,” but rather an alignment of the data model to fit the data, the effect still reduces the influence of bulk motion in SPECT. Both methods automate the correction process, minimizing user-dependent variability and resulting in more consistent and reproducible outcomes. SPECT acquisition physics and poor count statistics limit the degrees of freedom and number of corrections that can be reliably applied. Each correction method must be carefully designed to produce estimable parameters. In the implementation of MC Pro and rMC Pro, these parameters were tailored for conservative motion mitiga- tion. This guards against insufficient data, which may result in unintentional side effects of instability artifacts or poor reproducibility and repeatability. A central aspect of both methods is that they correct the most dominant motion that is visible in the data, and that corrections are applied to the entire FOV of each view. Therefore, turning on motion correction should be considered with the clinical application in mind and not as a default, as some regions may not benefit from motion correction. 16 White paper · Data-driven mitigation of SPECT motion-induced artifacts About the lead author Maximilian P. Reymann, MSc Maximilian Reymann is a nuclear imaging scientist in the Siemens Healthineers SPECT Research Department with specialized knowledge in motion correction for SPECT. He is writing his PhD thesis on the simulation, characterization, and improvement of motion correction methods in SPECT in cooperation with Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, and he has written several publications in this domain. White paper · Data-driven mitigation of SPECT motion-induced artifacts 17 References 1 Kyme AZ and Fulton RR. Motion estimation and correction in SPECT, PET and CT (in eng). Physics in Medicine and Biology. 2021; vol. 66 (no. 18). doi: 10.1088/1361-6560/ac093b. 2 Nakajima K, et al. Myocardial perfusion imaging with retrospective gating and integrated correction of attenuation, scatter, respiration, motion, and arrhythmia. J. Nucl. Cardiol. 2023/09/27 2023. doi: 10.1007/s12350-023-03374-5. 3 Cachovan M and Vija AH. One gate reconstruction. US Patent 10303849 Patent Appl. US15/314,481, 5/28/2019, 2019. [Online]. Available: https://patents.google.com/ patent/US10303849B2/en?oq=10303849 4 Cachovan M and Vija AH. Intra reconstruction motion correction. US Patent 10304219 Patent Appl. 15/315,313, 5/28/2019, 2019. [Online]. Available: https:// patents.google.com/patent/US10304219B2/en?oq=10304219 5 Sanders JC III and Vija AH. Respiratory motion estimation in projection domain in nuclear medical imaging. US Patent 10398382 Patent Appl. US15/721,166, 9/3/2019, 2019. [Online]. Available: https://patents.google.com/patent/US10398382B2/ en?oq=10398382 6 Sanders JC III and Vija AH. Data-driven surrogate respiratory signal generation for medical imaging. US Patent 10292671 Patent Appl. US15/567,492, 5/21/2019, 2019. [Online]. Available: https://patents.google.com/patent/US10292671B2/ en?oq=10292671 7 Sanders JC III, Ritt P, Kuwert T, Vija AH, and Maier AK. Fully automated data-driven respiratory signal extraction from spect images using laplacian eigenmaps. IEEE Trans. Med. Imaging. 2016 doi: doi: 10.1109/TMI.2016.2576899. 8 Vija AH. Extracting application dependent extra modal information from an anatomical imaging modality for use in reconstruction of functional imaging data. United States Patent Appl. 9332907. 2016. [Online]. Available: http://www. freepatentsonline.com/9332907.html 9 Vija AH and Cachovan M. Automated motion correction in quantitative spect reconstruction: a feasibility study of a method framework applied first to bone imaging. Journal of Nuclear Medicine. vol. 58, no. supplement 1, pp. 703-703, 2017. 10 Vija AH and Ding X. Numerical and visual assessment of fully automated method of reducing tomographic inconsistency in quantitative SPECT reconstruction of clinical data. Journal of Nuclear Medicine. vol. 61, no. supplement 1, pp. 1461-1461, 2020. 11 Vija AH and Yahil A. Method and apparatus for using image cumulative distribution function for tomographic reconstruction quality control. US Patent 8674315 Patent Appl. 12/944,871, 3/18/2014, 2013. [Online]. Available: https://patents.google.com/ patent/US8674315B2/en?oq=8674315 12 Ma J and Vija AH. Reconstruction quality assessment with local non-uniformity in nuclear imaging. US Patent 10210635 Patent Appl. US15/587,266, 2/19/2019, 2019. [Online]. Available: https://patents.google.com/patent/US10210635B2/ en?oq=10210635 18 White paper · Data-driven mitigation of SPECT motion-induced artifacts Trademarks and service marks used in this material Motion correction and respiratory motion correction are are property of Siemens Medical Solutions USA or offered on Symbia Pro.specta SPECT/CT systems. Siemens Healthineers AG. All other company, brand, product, and service names may be trademarks or Symbia Pro.specta and its features are not commercially registered trademarks of their respective holders. available in all countries. Future availability cannot be Please contact your local Siemens Healthineers sales guaranteed. representative for the most current information or contact one of the addresses listed below. All comparative claims derived from competitive data at the time of printing. Data on file. Siemens Healthineers reserves the right to modify the design and specifications contained herein without prior notice. As is generally true for technical specifications, the data contained herein varies within defined tolerances. Some configurations are optional. Product performance depends on the choice of system configuration. Note: Original images always lose a certain amount of detail when reproduced. “Siemens Healthineers” is considered a brand name. Its use is not intended to represent the legal entity to which this product is registered. All photographs © 2025 Siemens Healthineers AG. All rights reserved. Siemens Healthineers Headquarters Published by Siemens Healthineers AG Siemens Medical Solutions USA, Inc. Siemensstr. 3 Molecular Imaging 91301 Forchheim 2501 N. Barrington Road Germany Hoffman Estates, IL 60192 Phone: +49 9191 18-0 USA siemens-healthineers.com Phone: +1 847 304-7700 siemens-healthineers.com/mi MI-6627 TA.JV ∙ PDF ONLY ∙ © Siemens Healthineers AG, 01.2025