Factors influencing image quality and quantification: optimization of PET parameters

Transcription

1 Factors influencing image quality and quantification: optimization of PET parameters Ronald Boellaard Department of Radiology & Nuclear Medicine, VU University Medical Center, Amsterdam Department of Nuclear Medicine & Mol.Imaging, University Medical Center Groningen

2 Molecular Imaging with (Q)PET (with CT and/or MR) Different radiopharmaceuticals to image different metabolic pathways Diagnosis and staging Restaging Biological Characterization Response Evaluation Therapy

3 Optimization Task: what is the main purpose of the imaging procedure: Lesion detection Quantification Cross-sectional of longitudinal Cost function / metric used for optimisation: Lesion detection contrast noise ratio Quantification accuracy & precision -multicenter?

4 Why quantification? To reduce observer variability To provide a continuous scale / read-out. Quantification interpretation, it simply provides a quantitative metric for the imaging biomarker of interest

5

6 Data analysis methods Full kinetic analysis (using plasma input) Often requires invasive arterial sampling Plasma input processing is sensitive to errors and noise In practice often only Ki or V T (= free, non-spec+spec ) can be obtained reliably Complex, but more accurate, less assumptions Least or not feasible in clinical settings Standardized uptake value (SUV) or uptake ratio (SUVR) Simple, can be applied in the clinic More assumptions, can be misleading in some cases Needs validation & standardisation

7 Standardized Uptake Value SUV TBW = ct [ kbq / ml] Dose[ MBq ]/ weight[ kg] SUV is activity concentration ratio Weight is sometimes replaced by BSA, LBM (SUL), BMI

8 Boellaard et al, JNM 2004 Standard Uptake Value If tracer uptake reaches a constant value: SUV TBW = c t [ kbq / ml] Dose[ MBq ]/ weight[ kg] Instead of using total body weight (TBW) - lean body mass - body surface area can be used and are more appropriate (better correlation with glucose metabolic rate)

9 Boellaard et al, JNM 2004 Standard Uptake Value If tracer uptake reaches a constant value: SUV TBW = c t [ kbq / ml] Dose[ MBq ]/ weight[ kg] Dose = activity at time of study, i.e. it should be corrected for decay to start of PET study: - Essay dose (act.) at essay (calibration) time - Dose (act.) at time of administration - Start of PET of PET/CT study

10 Use of SUV in response assessment studies Absolute SUV: -Patient eligibility -Patient stratification -Lesion selection (PERCIST) -Residual SUV Relative of % SUV changes -% change of the same lesions (EORTC) -% change of the (5) hottest lesions per scan + ΔSUV=0.9 (PERCIST) For all applications absolute SUV and SUV changes are used

11 PET imaging / SUV uncertainties R. Boellaard 2009, J Nucl Med Supplement Issue 50: 11S Technical factors Relative calibration between PET scanner and dose calibrator (10%) Residual activity in syringe (5%) Incorrect synchronization of clocks (10%) Injection vs calibration time (10%) Quality of administration (50%) Physics related factors Scan acquisition parameters (15%) Image reconstruction parameters (30%) Use of contrast agents (15%) ROI (50%) SUV TBW = ct [ kbq / ml] Dose[ MBq ]/ weight[ kg] Biological factors Uptake period (15%) Patient motion and breathing (30%) Blood glucose levels (15%) - Changes in blood clearance

12 Quantification depends on interplay between image characteristics (resolution an noise) and data analysis System spatial resolution TOF Reconstruction settings Weight Scan duration Activity Scanner sensitivity Reconstruction settings Noise Quantification Contrast recovery /resolution Data analysis Max, peak, mean SUV, MATV, TLG Software

13 Sequence of all steps determines final SUV results: accumulation of sources of error

14 Factors affecting SUV: biological factors Biological factor Blood glucose level Uptake period Patient motion/breathing Patient comfort Inflammation Effect Lower uptake or SUV with increasing blood glucose level Higher SUV at increasing time interval between injection and start of PET study Image artefacts in case mismatch in position between CT-AC and emission scan and (possibly) lower SUV due to respiratory motion (resolution loss) Patient stress and uncomfortable waiting conditions increase uptake of FDG in muscle and/or brown fat and may affect SUV quantification, also in case of low tumour-to-background ratios due to spillover. Inflammatory processes near or at the tumour results in a false positive increase of SUV.

15 Glu 200 mg% Glu 79 mg% Karoline Spaepen-Sigrid Stroobants Department of Nuclear Medicine University Hospital Gasthuisberg Leuven, Belgium

16 Lowe VJ et al. Optimum scanning protocol?for FDG-PET evaluation of pulmonary

17 Factors affecting SUV: technical/physical factors Factor Scan acquisition parameters (acquisition mode, scan duration, bed overlap, FDG dose) Image reconstruction settings (number of iterations, filters, matrix size, zoom factors) Effect Affect signal to noise ratio (SNR) of PET scan. Poorer SNR results in an upward bias of SUV. Affect convergence of iterative reconstructions and final spatial resolution (filter, matrix, zoom) of reconstructed image. Insufficient convergence and lower resolution results in lower SUV and increases partial volume effects. Moreover, insufficient convergence makes SUV more dependent on surrounding activity distributions. Region of interest (ROI) strategy to derive SUV Normalisation factor in SUV calculation Correction for serum glucose level in SUV calculation Use of contrast agents during CT-AC Higher or lower SUV depending on size and type of ROI used. SUV outcomes are numerically different when using body weight, body surface area or lean body mass as normalisation factor in the SUV equation Higher serum glucose levels will result in underestimation of SUV. Use of a serum glucose level correction in the SUV equation will thus result in different SUV outcomes. Can result in overestimation of attenuation and thus results in higher SUV (upward bias).

18 Effects of different number of OSEM iterations, as seen in the Netherlands, on SUV SUVmax = SUV 50%=

19 Iterative recon convergence MLEM Iterative recons slowly grows to the right estimate, but at the cost of increased image noise Problem: when to stop?

20 Iterative reconstruction: non-uniform convergence True image 20 iterations 100 iterations sinogram

21 Iterative reconstructions: convergence

22 Latest generation of PET: Time of Flight 1. Introduction 2. TOF-PET 3. TOF reconstruction 4. Image quality gain 5. Conclusions Improved energy resolution --> reduce scatter Small coincidence window --> reduce randoms Time-of-Flight PET capable --> improve image quality for same counts T = 4 ns T = 500 ps ~ 7.5 cm 22

23

24 Iterative reconstruction: non-uniform convergence True image 20 iterations 100 iterations sinogram

25 Resolution modeling

26 Reconstruction based PVC HD reconstruction Resolution modeling PSF reconstruction Correct for partial volume effect during reconstruction Include scanner s resolution characteristics within reconstruction

27 Iterative reconstruction with resolution modeling image space projection space Iteration N Update Current estimate Projection + SMOOTH Estimated Estimated projection Measured projection Compare (e.g. - or / ) Error image Backprojection Error projection

28 EANM QC Image Quality Siemens mct 30 s EANM STD HD (RM only) HD (RM+ToF) Ultra HD 180s 600s Ring or Gibbs artefacts

29 SUV max = % SUV peak = % SUV liver mean = % SUV liver max = % TLR max,max = % TLR max,mean = % MATV (A50%) = ml -13% Total IMG act = % Image based metrics may still depend on image quality and characteristics

30

31 Optimization of activity over time 80 s and 90 s: fixed activities EANM 2010: weight and scan duration based EANM 2015 & Carlier: weight, scan duration and system performance based

32 Intercomparison of images Effect of administered activity

33 Groot et al., EJNMMI Res.2013 Intercomparison of images Effect of adminstered activity Linear dosage = dose per kg Quadratic = dose per kg 2 Liver noise as index of image quality

34 Intercomparison of images Effect of adminstered activity Linear dose Quadratic dose 70 kg 95 kg 69 kg 96 kg

35

36 3 min 3.5 min 4 min Optimize activity based on: -Patient weight/bmi in combination with scan duration -PET/CT system performance

37

38 Entire chain of process determines quantitative result of an imaging biomaker Needs consistency of the execution of imaging procedure in longitudinal setting

39 Quantification depends on interplay between image characteristics (resolution an noise) and data analysis System spatial resolution TOF Reconstruction settings Weight Scan duration Activity Scanner sensitivity Reconstruction settings Noise Quantification Contrast recovery /resolution Data analysis Max, peak, mean SUV, MATV, TLG Software

40 SUV requirements Accurate measurement of net injected dose Accurate measurement of weight, length of patient & (plasma glucose level) before scanning (patient weight changes over time!, use same device) Accurate calibration of PET scanner Accurate corrections of PET data Standard image acquisition procedures Equal time interval/acquisition parameters Standard reconstruction and filtering methods Equal image reconstruction settings resolution matching Standard data analysis methods (ROIs) Equal ROI methods Same SUV measures

41 FDG PET and PET/CT: EANM Procedure Guidelines for Tumour PET Imaging: version 2.0 Eur.J.Nucl.Med.Mol.Imag. 2015

42 * Boellaard et al. EJNMMI 2010, 2015 EANM Guideline & EARL accreditation* Main aim is to facilitate quantitative (SUV) evaluation of PET/CT examinations within multi-center studies. Aurthor suggest that same procedure is beneficial for quantitative clinical studies. However, there is room for other choices, in particular when performing visual intepretations only.

43 Development of guidelines and validation of Q-PET/CT as imaging biomarker requires harmonization: - repeatability = precision on a single system - reproducibility = precision across systems and sites

44 PET imaging / SUV uncertainties Technical factors Relative calibration between PET scanner and dose calibrator (10%) Residual activity in syringe (5%) Incorrect synchronization of clocks (10%) Injection vs calibration time (10%) Quality of administration (50%) Physics related factors Scan acquisition parameters (15%) Image reconstruction parameters (30%) Use of contrast agents (15%) ROI (50%) SUV TBW = ct [ kbq / ml] Dose[ MBq ]/ weight[ kg] Biological factors Uptake period (15%) Patient motion and breathing (30%) Blood glucose levels (15%) R. Boellaard 2009, J Nucl Med Supplement Issue 50: 11S

45 The EANM guideline for FDG PET and PET/CT provides recommendations for: Minimizing physiological or biological effects by patient preparation guidelines Procedures to ensure accurate FDG administration Matching of PET study statistics ( image quality ) by prescribing FDG dosage as function of patient weight, type of scanner, acquisition mode and scan duration Matching of image resolution by specifying image reconstruction settings and providing activity concentration recovery coefficients specifications (QC experiment) Standardization of data analysis by prescribing region of interest strategies and SU measures Multi-center QC/QA procedures for PET and PET/CT scanners

46

47 IDEAL

48 Centered at 1.0 Realistic/EARL Higher SUVs are not necessarily better Not defined by lowest denominator!

49 Should we use PSF reconstructions? Lasnon et al. EJNMMI 2013 Most accurate PSF + SUVmean (VOI=3D-50%) Most accurate No PSF + SUVmax Note that simple SUVmean & 3D 50% VOIs only perform well: - Simple phantoms - No tracer uptake heterogeneity - Good scan statistics None of these characteristics are met in clinical practice. (Cheebsumon et al. JNM 2011, EJNMMI 2011)

50 Why do we use SUVmax? SUVmax suffers from upward bias due to noise (Boellaard et al, JNM 1996, 2011, Lodge et al, JNM 2011) poor reproducibility and accuracy for PSF (HD) reconstructions (Tong, IEEE TNS 2011, Rahmim et al. MedPhys 2013, Lasnon et al. EJNMMI 2013) Despite these limitions: May represent metabolically most active part of tumor VOIs are not standardized simple isocontour work only well for simple phantoms CT and PET based manual segmentation suffer from observer variability CT based segmentation may suffer from CT-PET alignment issues PET based automated delineation methods: variability of methods variability in implementation of same method performance depend strongly on underlying image characteristics (Cheebsumon et al. JNM2011, EJNMMI 2011) cannot deal well with tracer uptake heterogeneity SUVmax is de facto the field standard (Rausch et al. NukMed 2014) Therefore, need to optimize image quantification for use of SUVmax

51 EARL HD/PSF For now: 2 reconstructions Future: vendor provided solutions to be announced/discussed in Gothenburg

52 The EARL FDG-PET/CT Accreditation Programme & Guideline Developments: Results of more than 65 Successfully Accredited Sites and Future Perspectives

53 Multi-center QC and calibration Daily QC conform standard procedure of system / manufacturer Calibration QC using (cylindrical) phantom (15-30cm diameter) Adjusted NEMA NU Image Quality procedure/measurement to measure recovery coefficients as function of sphere size (= effective image resolution ) CT-QC cf recommendations of ESR/national law Misc. QC (e.g. for scales, alignment etc)

54 Multi-center harmonization of quantification Calibration QC PET/CT and DC Results at first test accredited sites GE Philips Siemens GE Philips Siemens GE Philips Siemens ~5% outside specs Comparable calibration accuracy and SUV recovery among sites and vendors is feasible (n=>65)

55 Multi-center harmonization of quantification Image Quality % SUV recovery [BQML] Austrian findings EARL findings, at 1st intake 1.60 SUV Max RC - all vendors SUV MAX RC Sphere volume (ml)

56 Multi-center harmonization of quantification Image Quality % SUV recovery [BQML] Results at first test accredited sites SUV Max RC - all vendors 1.6 SUV MAX RC - all vendors SUV MAX RC SUV MAX RC Sphere volume (ml) Sphere volume (ml) Comparable SUV recovery among sites and vendors is feasible (n=>65)

57 Future directions (EARL v2.0) SUVmax suffers from: upward bias due to noise (Boellaard et al, JNM 1996, 2011, Lodge et al, JNM 2011, Lasnon EJNMMI 2013) poor reproducibility and accuracy for PSF (HD) reconstruction (Boellaard et al., JNM 2011, Tong, IEEE TNS 2011, Lasnon et al. JNM 2013) 1. Explore use of SUVpeak: 1ml spherical VOI located at highest average value good surrogate for SUVmax almost no observer variability less sensitive to scanner performance differences (Makris et al. JNM 2014) 2. Explore implementation of EARL compliant acq/recon protocols by vendors Positive feedback from and ongoing discussions with GE, Philips and Siemens Explore strategy proposed by Lasnon et al. EJNMMI nd recon or post-recon filter after PSF recon Optimize multicenter quantification and lesion detectability 3. Include measure and upper limit for noise within the IQ-QC experiments optimization of administered activity taking also performance of PET/CT system into account.

58 The future of molecular imaging New analysis methods/other uptake parameters

59 Thanks for your attention