EQ•PET: Achieving NEMA-referenced SUV across technologies

SIEMENS White Paper EQ•PET: Achieving NEMA- referenced SUV Across Technologies Matthew Kelly, PhD, Siemens Healthcare Sector Answers for life. Table of Contents Introduction 1 Case Study 1 – Cross-Scanner Response Assessment 2 Clinical Example 2 Case Study 2 – Multi-Center Clinical Trials 4 Clinical Example 4 Case Study 3 – Inter-Site SUV Thresholds 6 Conclusions 6 Appendix A: EQ•PET Parameter Determination 7 Test Protocol 7 Target Protocol 7 Recovery Coefficient Measurement 7 EQ•PET Parameter Optimization 8 Appendix B: Computing EQ•PET Parameters in syngo.via 9 Example Dataset 9 EQ•PET Parameter Optimization 9 Appendix C: Gaussian Smoothing Methodology 11 Kernel Construction 11 About the Author 12 References 12 Introduction PET is a valuable tool for helping diagnose, stage and monitor EQ•PET achieves this without requiring the clinical site to modify cancer, as well as enabling clinicians to quantify active disease and their reconstruction protocol or reconstruct additional datasets. measure response to therapy. Accurate quantification aids clini- With EQ•PET, the clinician reads from the original patient image, cians in benchmarking disease and identifying effective therapies reconstructed with their preferred protocol to maximize image earlier in the treatment cycle, thus improving the efficiency and quality and detectability. SUV is harmonized using an EQ•PET efficacy of patient care. parameter selected to align contrast recovery between scanners and reconstructions (Figure 1), relative to a reference such as the The Standardized Uptake Value (SUV) is the most widely-used EANM specification3. metric for quantifying radiotracer uptake in tumors, providing normalization for differences in patient size, body composition This white paper presents three clinical use cases for SUV and injected dose; however, differences in scanner hardware and harmonization along with an approach for determining the EQ•PET reconstruction protocol can introduce clinically significant variation parameter for the scanner model and reconstruction protocol. The in PET quantification that are not addressed by SUV alone1. clinical use cases presented are: EQ•PET is a new reference-based quantification technology within • Cross-scanner response assessment for patients imaged syngo®.via that provides clinicians with harmonized SUVs across with different PET/CT systems; patient scans, even if acquired on different scanners or recon- • Multi-center clinical trials that require strict alignment of structed with different protocols2. acquisition protocol and quantitative performance of scanners; • Inter-site SUV thresholds to facilitate the exchange and adoption of SUV-based protocols between clinical sites Patient PET EQ. PET EQ . PET Parameter ............................ ............................... .............................. NEMA IQ Recovery Recovery.eq Figure 1. EQ•PET harmonizes SUVs across different scanners and reconstructions by applying a phantom-derived reference-based EQ•PET parameter optimized to align contrast recovery coefficients. 1 Case Study 1 – Cross-Scanner Response Assessment Challenge: Assess treatment response in This improved comparability across reconstructions with EQ•PET is patients quantitatively, even if imaged on also seen with SUVpeak (Table 2). While the PERCIST-based response different PET/CT systems. classification is not affected in this example, a difference in SUVpeak of +24% versus -3% has the potential to impact a clinician’s assess- ment of treatment effect. PET/CT imaging is used clinically to assess a cancer patient’s A prospective evaluation of the impact of applying a phantom- response to treatment. While dramatic disease progression or treat- derived parameter to align quantification found that it allowed for ment response can often be reliably determined from a qualitative a reliable pre- and post-therapy evaluation when using different review of the images, more subtle changes require quantitative generation PET systems1. assessment4. Furthermore, quantitative assessment enables objec- tive evaluation of change, with standardized response criteria such as PERCIST4, improving inter-reader agreement. Post-RT scan Change in SUVmax (%) Quantitative response assessment is typically performed using Iterative HD•PET HD•PET.eq SUV, which normalizes for differences in dose injected and patient Iterative +10% (SMD) +92% (PMD) +9% (SMD) weight to facilitate inter-scan comparison. Despite this normaliza- Pre-RT scan HD•PET tion, differences in scanner model and reconstruction can still have -28% (PMR) +25% (PMD) n/a a clinically significant impact on SUV1. HD•PET.eq +4% (SMD) n/a +4% (SMD) Table 1. Percentage change in SUVmax between pre- and post-RT scans for EQ•PET quantification, in combination with a standardized different reconstruction protocols. EORTC response classification is denoted imaging protocol, allows a physician to assess treatment response by (PMD) progressive metabolic disease, (SMD) stable metabolic disease in patients quantitatively, even if the patient’s scans were acquired and (PMR) partial metabolic response. on different systems or reconstructed differently. n/a = not applicable Clinical Example EQ•PET is mitigating the impact of the reconstruction protocol which could impact patient management. The following lung cancer patient (Figure 2) received 2 cycles of chemotherapy, including granulocyte stimulating factors, prior to radiotherapy. The first PET/CT scan was performed prior to radio- therapy with the second 8 weeks later. Change in SUVpeak (%) Post-RT scan Iterative HD•PET HD•PET.eq The pre-RT scan was reconstructed using Iterative (OSEM) with 4 Iterative -3% (SMD) +24% (SMD) +1% (SMD) iterations, 8 subsets and a 5 mm FWHM Gaussian post filter. The Pre-RT scan HD•PET post-RT scan was reconstructed using HD•PET (PSF) with 3 itera- -23% (SMD) -2% (SMD) n/a tions, 21 subsets and no post filter (Figure 2). HD•PET.eq -7% (SMD) n/a -3% (SMD) Table 2. Percentage change in SUVpeak between pre- and post-RT scans Using the EORTC criteria5, the change in SUVmax between the for different reconstruction protocols. PERCIST response classification is two scans (3.74 to 7.17; +92%) indicates disease progression denoted by PMD, SMD and PMR. (Table 1). However, due to the difference in reconstruction, a confident assessment cannot be made. In fact, in an additional n/a = not applicable reconstruction of the post-RT scan with the pre-RT protocol, the EQ•PET is mitigating the impact of the reconstruction SUVmax measured for the lesion is 4.13 (+10%), indicating stable protocol which could impact patient management. disease according to the EORTC criteria. Using the appropriate EQ•PET parameter to align the HD•PET reconstruction with Iterative (7.0 mm FWHM), the SUVmax.eq measured for the same lesion on the post-RT HD•PET scan is 4.06 (+9%). EQ•PET, therefore, enables quantitatively comparable response assessment, despite the use of a more advanced recon- struction protocol with improved image quality in the post-RT scan. 2 PRE-RT (Iterative) Figure 2. Pre- and post-RT PET/ CT scans for lung cancer patient. A single lesion in the right lung is identified and the quantitative L1VOI2 PT assessment is dependent on the Max: 3.74 SUV-bw reconstruction method used. This Peak: 3.26 SUV-bw / Size: 1 cm3 dependency is minimized with Thresh: 1.87 SUV-bw / 50 % EQ•PET. Data courtesy of François Baclesse Cancer Centre, Caen, France. POST-RT (HD•PET) L1/013 PT Max: 7.17 SUV-bw Max.eq: 4.06 SUV-bw Peak: 4.03 SUV-bw / Size: 1 cm3 Peakeq: 3.30 SUV-bw / Size: 1 cm3 Thresh: 3.59 SUV-bw / 50 % 7.17 ( HD . PET ) PMD 4.13 (Iterative) 3.74 (Iterative) 4.06 (HD.PET.eq) SMD SUV max PMR Pre-RT Post-RT 3 Case Study 2 – Multi-Center Clinical Trials Challenge: Participate in multi-center clinical In each reconstruction, the SUVmax was measured for a small trials, without needing to modify established lesion in the left lung. When compared with the basic Iterative reconstruction protocols. reconstruction, both HD•PET and ultraHD•PET produce a clinically significant increase in SUVmax5 (Table 3). However, by applying the EQ•PET parameters necessary to align with the EANM specifica- tion2, this variability is reduced to within the reported test-retest Multi-center clinical trials facilitate the recruitment of a larger variability for 18F-FDG uptake in tumors6. number of subjects in a shorter time period. Medical imaging techniques, including PET/CT, are frequently used in clinical trials for patient stratification or as surrogate endpoints. SUVmax SUVmax.eq EQ•PET parameter The need for quantitative comparability in clinical trial imaging typically requires a form of site accreditation prior Iterative 4.04 3.80 3.3 mm to participation, such as that provided by the Society of HD•PET 6.37 (+58%) 3.45 (-9%) 6.5 mm Nuclear Medicine and Molecular Imaging’s Clinical Trials Network (CTN) or EANM Research Limited (EARL). As part UltraHD•PET 7.51 (+86%) 3.21 (-16%) 7.1 mm of this accreditation, each site must conform to a commonly Table 3. Effect of reconstruction on SUVmax and SUVmax.eq for the lung achievable standard for quantification. lesion shown in Figure 3. Percentage change relative to Iterative is shown in parentheses. EQ•PET parameters FWHMs required to align with the Typically, this will require the modification of a site’s reconstruction EANM specification are shown in the final column. protocol to conform with the trial protocol (usually leading to a loss of resolution) or, if this is undesirable, the reconstruction of EQ•PET is mitigating the impact of the reconstruction protocol which could an additional PET dataset for quantification. impact patient management. EQ•PET, in combination with a standardized imaging protocol, allows a site to adhere to the quantitative requirements for a multi-center clinical trial without having to reduce image quality or reconstruct and manage a second dataset. Clinical Example To demonstrate the quantitative impact of reconstruction, a lung cancer patient scan (Figure 3) has been reconstructed with three different protocols: • Iterative (OSEM) with 2 iterations, 24 subsets and a 5 mm FWHM Gaussian post filter • HD•PET (PSF) with 3 iterations, 24 subsets and a 4 mm FWHM Gaussian post filter • ultraHD•PET (PSF+TOF) with 3 iterations, 21 subsets and no post filter 4 Figure 3. Lung cancer patient Iterative reconstructed with Iterative, HD•PET, and UltraHD•PET. The SUVmax for a small lesion in the posterior left lung increases with the addition of advanced L1VOI1 PT reconstruction techniques. R Max: 4.04 SUV-bw Max.eq: 3.80 SUV-bw Data courtesy of University of Tennessee Medical Center, Knoxville, TN, USA. HD•PET L1VOI2 PT R Max: 6.37 SUV-bw Max.eq: 3.45 SUV-bw UltraHD•PET L1VOI3 PT R Max: 7.57 SUV-bw Max.eq: 3.21 SUV-bw 5 Case Study 3 – Conclusions Inter-Site SUV Thresholds EQ•PET provides clinicians with harmonized SUVs, allowing them to: Challenge: Use SUV-based thresholds to guide • Quantitatively assess treatment response in patients, patient management decisions, even if they even if imaged on different PET/CT systems • were defined on an older PET/CT system. Participate in multi-center clinical trials, without modifying their reconstruction protocols • Use SUV-based thresholds to inform patient management decisions, even if they were defined on an older PET/CT system PET/CT is an integral part of staging many cancer types and guides subsequent patient management decisions. Initial assessment of a suspicious lesion’s malignancy is commonly performed based on 18F-FDG uptake, with some sites using SUV-based thresholds to categorize lesions in terms of likelihood of malignancy. Further- more, SUV-based thresholds of change for classifying treatment response are also widely used (e.g., EORTC, PERCIST). SUV-based thresholds determined at one site may not be appli- cable to patient data acquired at other sites or using other scanner models or reconstruction protocols. EQ•PET, in combination with a standardized imaging protocol, facilitates the inter-site application of SUV-based thresholds for staging or response assessment. Consider the clinical example in Case Study 1. Based on a 25% SUVmax change threshold (EORTC), there is considerable variation in response classification with the different reconstruction combi- nations, even when the pre- and post-RT protocols are matched. This variation is reduced with EQ•PET. Given the clinical example in Case Study 2, it is clear that using a fixed SUV threshold to estimate likelihood of malignancy could result in different clinical decisions being made for a same patient depending on how it was reconstructed. This dependency on reconstruction can be reduced with reference-based EQ•PET. The results shown in the examples in this whitepaper are char- acteristic of a more comprehensive study2, which demonstrated a significant reduction in reconstruction-dependent variation for both SUVmax and SUVpeak. 6 Appendix A: EQ•PET Parameter Target Protocol Determination The target protocol used in this example is that specified in the EANM procedure guidelines3 for the maximum voxel value (Table 5). EQ•PET parameters can be selected to align PET quantification with lower recovery protocols. These include specifications published by international societies (e.g., EANM procedure guidelines), or EANM Sphere diameter (mm) Expected RC for max voxel existing PET protocols used within imaging centres. 10 0.38 The following illustrates one approach for selecting an appro- 13 0.63 priate harmonization parameter. This approach requires a NEMA 17 0.84 IQ phantom acquired and reconstructed using the same protocol used for clinical PET/CT studies. 22 0.89 28 0.95 Test Protocol 37 0.98 In this example, we used a NEMA IQ phantom filled with an 8:1 Table 5. Target recovery coefficient (RC) specifications for maximum voxel value sphere to background activity concentration ratio, acquired for used in this example.3 3 minutes on a Biograph mCT 64 with TrueV and reconstructed with ultraHD•PET (Table 4). The acquisition was repeated 3 times. Recovery Coefficient Measurement Property Value For each phantom acquisition, the recovery coefficient (R) for the Phantom activity 5.2 kBq.ml-1 (background) maximum voxel in each sphere was computed as follows: 41.6 kBq.ml-1 (spheres, 8:1) Acquisition duration 180 s (30x106 net trues) R =- Ameasured Scanner model Biograph mCT 64 with TrueV Atrue Reconstruction method UltraHD (PSF+TOF) 2i21s Convolution kernel 5 mm FWHM Gaussian where measured is the maximum activity concentration (Bq.ml-1) Rows x columns 256x256 measured in an image voxel for a VOI corresponding to the sphere Pixel spacing insert, and Atrue is the true activity concentration (Bq.ml-1) in 3.182x3.182 mm the sphere.3 Slice thickness 2.027 mm Table 4. Phantom, acquisition and reconstruction properties used in this example. 7 EQ•PET Parameter Optimization A Conventional Recovery Coefficients 1.2 The RCs measured for each phantom acquisition were compared to those specified in the EANM procedure guidelines (Figure 4A). The RCs were then recomputed following the application of an 1 additional Gaussian smoothing filter (see Appendix C for details). The size of this additional filter was increased in steps of 0.1 mm FWHM until the mean absolute percentage difference between :0.8 the measured RCs and those specified in the EANM guideline was minimized (Figure 4B). The filter sizes required to minimize 0.6 the percentage difference for the acquisitions used in this example are shown in Table 6. Based on these acquisitions, the recom- mended EQ•PET parameter to align the UltraHD reconstruction 0.4 -- EANM Recovery coefficient protocol used in this example with the EANM guideline would Acquisition 1 be 6.8 mm FWHM. 0.2 Acquisition 2 - Acquisition 3 Phantom acquisition EQ•PET parameter O FWHM (mm) 10 20 30 40 O Acquisition 1 Sphere diameter (mm) 6.7 Acquisition 2 7.1 Acquisition 3 6.6 B Recovery Coefficients with EQ•PET Harmonization Mean 6.8 1.2 Table 6. EQ•PET normalization parameters computed for the 3 phantom acquisitions used in this example. 1 0.8 .6 0.4 Recovery coefficient -- EANM 0.2 Acquisition 1 Acquisition 2 - Acquisition 3 O 10 20 30 40 O Sphere diameter (mm) Figure 4. (A) RCs measured for the 3 acquisitions of the NEMA phantom relative to those specified in the EANM guidelines.3 (B) RCs measured for the same 3 acquisitions following application of the EANM-matching EQ•PET harmonization parameter. 8 Appendix B: Computing EQ•PET EQ•PET Parameter Optimization Parameters in syngo.via The steps used to compute the optimal EQ parameter for the phantom dataset described above are as follows: This section describes how syngo.via can be used to compute 1. The target activity concentration for each of the phantom hot the optimal EQ parameter size for a given scanner model and spheres (column ‘EANM target AC’ in Table 8) was computed reconstruction protocol. by multiplying the true decay-corrected activity concentration (column ‘True AC’ in Table 8) by the corresponding EANM target Example Dataset recovery coefficient (column ‘EANM target recovery’ in Table 8). 2. Create a spreadsheet (e.g., in Microsoft Excel™) to compute the In this example, a NEMA IQ phantom was prepared and recon- mean absolute percentage difference in activity concentrations structed as described in Table 7. for the Max.eq values with a given EQ parameter versus the target values computed in Step 1 (Table 8). 3. The NEMA IQ phantom dataset is loaded into syngo.via Property Value MMOncology. True AC in sphere 18.70 kBq.ml-1 4. The units are set to Bq.ml-1 from the units entry in the bottom Biograph 6 with TrueV right corner menu. Scanner model 5. Each of the hot spheres is segmented using one of the available Reconstruction method HD•PET (PSF) 3i21s PET segmentation tools accessed via the top right corner menu Convolution kernel All pass (0 mm) (e.g., VOI Isocontour) (Figure 5). Rows x columns 168x168 6. Ensure Max.eq is displayed in the findings evaluation text. This is Pixel spacing set in the Segmentation Properties dialogue (Figure 6) accessed 4.073x4.073 mm by right-clicking on the chosen segmentation tool item in the Slice thickness 5.0 mm top right corner menu. Table 7. Phantom preparation and reconstruction protocol used in 7. Find the EQ parameter that minimizes the mean absolute this example. percentage difference in activity concentrations for the Max. eq values versus the target values computed. This could be done by entering EQ parameter sizes in 0.1 mm increments until the minimum is found. 8. For the dataset used in this example, the optimal EQ parameter size was 6.5 mm with a mean absolute percentage difference of 4.43 % (Table 8). A Figure 5. Segmentation of the hot LGVO11 spheres in the NEMA IQ phantom Max: 22.86 kBq/ml Max.eq: 18.59 kBq/ml with syngo.via MMOncology. L1VOI1 Max: 10.73 kBq/ml Max.eq: 6.43 kBq/ml L5VOI1 Max: 22.04 kBq/ml Max.eq: 17.65 kBq/ml -R L2VOI1 Max: 19.50 kBq/ml Max.eq: 11.76 kBq/ml L4VOI1 Max: 23.89 kBq/ml Max.eq: 18.01 kBq/ml L3VOI1 Max: 23.13 kBq/ml Max.eq: 16.73 kBq/ml 9 PET/SPECT Segmentation Properties ? RECIST 1.1 v Figure 6. Turning on display of Display the following evaluations on PET/SPECT data: Max.eq via the Segmentation 'auto Properties dialogue (left) and setting Max (Maximum) the EQ parameter via the Quantifica- Max.eq (Maximum Equivalent) tion Parameters blind (right). auto Peak ( Highest average 1 cm 3 sphere ) Quantification Parameters Peak.eq (Highest average 1cm3 sphere equivalent) Sex: O Male Female . Other Mean (Average) Height: cm Mean.eq (Average Equivalent) Weight: 10 kg SD (Standard Deviation) Isotope: F-18 SD.eq (Standard Deviation Equivalent) Half-life: 109.77 Minutes Min (Minimum) Dose: 19.56 MBq Volume Injection Date/Time: 27/11/2013 17:40:00 Thresh (Threshold of VOI Max) EQ Filter: 6.5] mm Reference Regions OK Cancel EQ Parameter = 0 mm EQ Parameter = 6.5 mm Sphere EANM target True AC EANM target Measured AC % diff. vs. Measured AC % diff. vs. diameter recovery (kBq.ml-1) AC (kBq.ml-1) (kBq.ml-1) EANM AC (kBq.ml-1) EANM AC 10 mm 0.38 18.70 7.11 10.73 50.9 % 6.43 -9.6 % 13 mm 0.63 18.70 11.78 19.50 65.6 % 11.76 -0.2 % 17 mm 0.84 18.70 15.71 23.13 47.2 % 16.73 6.5 % 22 mm 0.89 18.70 16.64 23.89 43.6 % 18.01 8.2 % 28 mm 0.95 18.70 17.77 22.04 24.0 % 17.65 -0.7 % 37 mm 0.98 18.70 18.33 22.86 24.7 % 18.59 1.4 % Mean absolute % difference vs. EANM target: 42.7 % 4.4 % Table 8. Target and measured activity concentrations (AC) for the phantom dataset used in this example. An EQ parameter of 6.5 mm provided the best alignment (smallest mean absolute percentage diffence) with the EANM target recovery coefficients. 10 Appendix C: Gaussian Smoothing Where dx, dy and dz are the distances (in units of number of voxels) Methodology of the center of the kernel voxel from the center of the kernel. N is the sum of all voxel values in the kernel to ensure the final values sum to 1. σx, σy and σz specify the sigma of the Gaussian kernel Gaussian smoothing for EQ•PET in syngo.via is performed using in each dimension (in units of number of voxels). a direct convolution operation with the kernel constructed as described below. The size or extent, g, of the kernel (in units of number of voxels) is computed from the Gaussian FWHM (in units of number of voxels) Kernel Construction in each dimension as follows: The function for calculating each element’s value in the Gaussian T Y $ FWHMx Kernel matrix can be defined as: gx T Y gy round up to the next odd number 2 FWHMy = + 1 gz FWHMz 2 1 1 @ + + # dx 2 dy 2 dz $ f@dx , dy , dz ,# 2 2 σx 2 σy 2 = e σz N The kernel extent in each dimension must be an odd number of voxels to ensure it is symmetric around the central voxel. FWHM g ----- ......................................................................................................................................................................................... --- - - - - - - - - -- - - Figure 7. Relationship between FWHM, sigma (σ) and kernel size or extent (g) for a Gaussian kernel. The intensity profile (blue) is shown for a single dimension of the kernel (grey grid). During the convolution, any image voxels outside the extent of the kernel are ignored. 11 About the Author References Dr. Matthew Kelly received a B.Sc. (hons) in Applied Neuroscience 1 Lasnon C, Desmonts C, Quak E, et al. Harmonizing SUVs from the University of Manchester in 2001, and then completed in multi-center trials when using different generation PET a Ph.D. in Pharmacology at the University of Cambridge in 2005. systems: prospective validation in non-small cell lung cancer He has worked in medical imaging since 2006 at the University of patients. Eur J Nucl Med Mol Imaging. 2013;40:985. Oxford and joined the Science and Technology Team of Siemens Molecular Imaging in 2007. 2 Kelly M, Declerck J. SUVref: reducing reconstruction- dependent variation in PET SUV. Eur J Nucl Med Mol Imaging Res. 2011;1:16. 3 Boellaard R, O’Doherty MJ, Weber WA, et al. FDG PET and PET/CT: EANM procedure guidelines for tumor PET imaging: version 1.0. Eur J Nucl Med Mol Imaging. 2010;37:181. 4 Wahl R, Jacene H, Kasamon Y, Lodge M. From RECIST to PERCIST: Evolving Considerations for PET Response Criteria in Solid Tumors. J Nucl Med. 2009;50:122S. 5 Young H, Baum R, Cremerius U, et al. Measurement of Clinical and Subclinical Tumor Response Usinf [18F]-fluorode- oxyglucose and Positron Emission Tomography: Review and 1999 EORTC Recommendations. Eur J Cancer. 1999;35:1773. 6 de Langen A, Vincent A, Velasquez L, et al. Repeatability of 18F-FDG Uptake Measurements in Tumors: A Metaanalysis. J Nucl Med. 2012;53:701. 7 National Electrical Manufacturer’s Association. NEMA Standards Publication NU 2-2007. Performance Measurements of Positron Emission Tomographs. NEMA; 2007. 12 Trademarks and service marks used in this material are property of Siemens Medical Solutions USA or Siemens AG. Siemens Medical Solutions USA, Inc. © Siemens Medical Solutions USA, Inc. All rights reserved. All photographs © Siemens Medical Solutions, USA, Inc. All rights reserved. Note: Original images always lose a certain amount of detail when reproduced. Global Business Unit Siemens Medical Solutions USA, Inc. Molecular Imaging 2501 N. Barrington Road Hoffman Estates, IL 60192-2061 USA Telephone: +1 847 304 7700 www.siemens.com/mi Global Siemens Headquarters Global Siemens Headquarters Address of legal manufacturer Siemens AG Healthcare Headquarters Siemens Medical Solutions USA, Inc. Wittelsbacherplatz 2 Siemens AG Molecular Imaging 80333 Munich Healthcare Sector 2501 N. Barrington Road Germany Henkestrasse 127 Hoffman Estates, IL 60192-2061 91052 Erlangen USA Germany Telephone: +1 847 304 7700 Telephone: +49 9131 84-0 www.usa.siemens.com/mi www.siemens.com/healthcare Order No. A91MI-10410-1T-7600 | All rights reserved | MI-1345.TM.JV.TW.1500 | © 05.2014, Siemens AG www.siemens.com/mi