Survey
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
Rigshospitalet, Copenhagen PET (and PET/CT, and SPECT/CT) Medical Imaging Systems DTU, October 2010 PET Søren Holm Entrance 39 Ansv. fysiker, lic. scient PET- and Cyclotron Unit, Rigshospitalet $2010 1 $2010 Photo: Flying Professor Andreas Kjær 2 Overview (0): Thanks to: GE Medical Systems CPS Innovation (Bernard Bendriem) Siemens Medical Solutions Philips Medico Different kinds of tomography – PET in particular Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) Impact (www.impactscan.org) $2010 3 (T)CT ECT $2010 SPECT 4 CT PET PET Computer(ized) X-ray ON from: $2010 attenuation of photons distribution of tracer attenuation of photons distribution of tracer structure / anatomy function / physiology structure / anatomy function / physiology 5 $2010 6 1 Radiology Nuclear Medicine ”Noise” in images depend on counts 5M 5k Ken Krowlton, Domino Portraits, 1982 #dots recorded determine what details can be seen in image $2010 And for the ignorants: this is Groucho Marx CT 7 distribution of tracer structure / anatomy function / physiology Map of Nuclides $2010 9 $2010 Positron #protoner for 8 PET attenuation of photons P $2010 10 PET is really ART: Annihilation Radiation Tomography α 511 keV 1955 β+ β+ positrondecay e- stable nuclides β− 511 keV electrondecay $2010 #neutroner E = mc2 1905 11 $2010 Result: Two photons on an almost straight line that nearly contains the point of decay 12 2 PET radiation detection Positron Emission Tomography (?) or ART We never detect the positrons directly, only the annihilation photons Annihilation A PET scanner counts coincidences Radiation Tomography detector d1 Coincidence window ee+ Tube or Line Of Response (LOR) 13 yes ∆t ~4-12 ns t2 detector d2 coincidence ∆t time 1 Coincidences $2010 |t1-t2|<∆t? d1 d2 Inhalation of radioactive β + gas 1991 t1 2 3 4 5 $2010 14 Map of Nuclides Time of Flight (TOF) PET – more than just a coincidence... tried and abandoned in ancient PET times (~1985) idea commercially relaunched in 2006 by Philips measure time difference between arrival at the 2 detectors ∆t = 600 ps = 0.6 ns Annihilation point can be determined with an accuracy of 0.6 ns × 30 cm/ns / 2 ⇒ 9 cm (why “/2” ?) det1 ee+ det2 Hence, more information available for image formation Hypothesis: better image quality, or shorter scanning time, or less injected activity $2010 15 $2010 16 Positron Emitting Isotopes …magnified Isotope Half-Life Production Carbon-11 20.5 min 14N(p,α α)11C Nitrogen-13 10.0 min 16O(p,α) α)13N Oxygen-15 2.1 min 14N(d,n)15O Fluorine-18 110 min 18O(p,n)18F Gallium-68 68 min Daughter of Ge-68 (271days) Rubidium-82 1.3 min Daughter of Sr-82 (25days) α α) (F-), 20Ne(d,α α)18F (F2) • Small elements (C,N,O,F) allow “real” biochemistry • Short half-lives make tracer production an integral part of PET $2010 17 $2010 3 (small) Cyclotron Cyclotrons Ion Source - Production of H RF + magnetic field Ion acceleration Carbon foil Electron stripping - H ⇒ H+ (protons!) protons do irradiate a Scanditronix 32 MeV GE Minitrace 11 MeV Target $2010 19 $2010 20 Production of 18F Nuclear Reactions Z 17F 18O (p,n) 18F 14N (d,n) 15O 14N (p,α α) 11C 16O (p,α α) 13N 18F 64.49s β+ 109.77m 18O(p,n)18F 19F 20.3 ms post FDG synthesis ≈ 20 GBq / 20 µA / 2h irr 15O 16O 17O 18O 122.24s 99.762 0.038 0.200 product H218O (150 euro/g) 18F- 17O 18O 99.64 0.36 0.038 0.200 12C 13C 14C 98.89 1.11 5736 a also used: 20Ne(d,α α)18F N β+ ≈ 0.7 GBq/µA sat 21 $2010 target material product Ne (+ F2) gas (1 euro/liter) 18F-F 2 $2010 Overview (1): 19F 20.3 ms 16O $2010 18F 109.77 96.9 m % β+ 99.762 β+ 99.8 % target material 15N 14N 9.96 m 20.3 m YIELD ≈ 2 GBq / µA sat 96.9 % β+ 99.9 % β+ 13N 11C 20Ne 90.48 Cheap, but... low specific activity 22 Principle of image formation! Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) Projection 23 $2010 Backprojection 24 4 $2010 25 $2010 26 27 $2010 28 Eksempel på FBP Filtered Back Projection af røntgendata (CT) $2010 Iteration principle image space image estimate old forward projection model !!! Iterative methods image space projection space Iteration 1 projection space estimated projections projection measured projections Estimated projection new UPDATE COMPARE image space error projection space error Measured projection Current estimate Compare (e.g. – or / ) Update Error image backprojection $2010 29 $2010 backprojection Error projection 30 5 Iterative methods image space projection Current estimate Iterative methods Iteration 2 projection space image space projection Estimated Estimated projection projection Measured projection Current estimate Compare (e.g. - or / ) Update projection space Iteration N Estimated Estimated Estimated Estimated projection projection projection projection Measured projection Compare (e.g. - or / ) Update Error image backprojection Error projection Error image $2010 31 Error projection $2010 Iterative Recon versus FBP 32 Overview (2): FBP Methods Iterative Methods Implementation: Reconstruction speed: Imbedded corrections: Simple Very fast None Complex Slow with ML-EM, fast with OSEM anything you can model Low-count image: Reconstruction artifacts: Noisy Streak/spill artifacts Less noise None, regular activity distribution FBP reconstructed image backprojection Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) OSEM reconstructed image $2010 33 $2010 34 (IAEA meeting 1959) Positron scanner (not PET) (Brain tumor scanner at MGH 1955- ) Positrocephalogram (PCG) of patient with meningioma PCG: A ”routine” examination! $2010 35 $2010 36 6 Anger camera (Rev.Sci.Instr. 1957) Gamma camera for positron imaging: NOT a recent invention $2010 37 $2010 38 Gamma camera principle position sensitive photon-detector Principles proposed by Hal Anger 1959 ||||||||||||||||||||||||||||||||||||||||||||||||||| PARALLEL HOLE COLLIMATOR used for planar imaging or tomography (SPECT) $2010 39 $2010 40 PET 1966 - 1968 PET Positome I 1966 32 detectors Advance 1993 12096 detectors HRRT 2002 119,808 detectors Therascan 1982 256 detectors $2010 41 $2010 $2002 42 7 Special Brain scanner (HRRT) Overview (3): Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) 119.808 Detector elements $2010 43 Coincidence Projections $2010 44 Data-Sorting Sinogram Counts Projection Projection +90° Sinogram Angle Detector ring 0° Position -90° Coincidence events determine sampling paths (Lines Of Response) Parallel Lines Of Response are sorted into a row of count numbers (Projection), representing the number of 511 keV photon pairs detected. 45 $2010 Each projection is entered as a row into a sinogram. A sinogram is an array which stores the number of coincidence events for each detector position and each angle. $2010 46 Geometric correction Sinogram Interpolation needed before reconstruction the center of the field of view. The sampled LORs are not equidistant, but become closer towards the edges. r y Real Space φ x Project Radon transform The projections must be resampled prior to or during reconstruction Point in center: straight line Off-center: sine curve, amplitude proportional to radius $2010 47 $2010 48 8 Image Reconstruction The Collimator in ring PET Image slice Sinogram $2010 49 $2010 Static Frame Acquisition 50 Dynamic Frame Acquisition Set of slices in a single FOV No patient table motion Brain scan, tumor spot, heart examinations (myocardial perfusion, metabolism and viability) Set of slices over time in a single FOV (like a movie) No patient table motion Start data acquisition as the tracer is administered Quantitative flow study (blood or other fluid): brain or heart Cerebral dynamic PET scan Tracer is 18F-labeled D2 dopaminergic receptor ligand Finding: caudate and putamen normal tracer uptake Adjacent brain PET static slices Gated Acquisition time Gated Acquisition Set of slices over time in a single FOV (like a movie) No patient table motion Heart imaging without cardiac motion blurring Myocardial perfusion, metabolism and viability R-wave R-wave ECG Heart cycle Electrocardiogram Gated slices of a cardiac PET scan time 9 List Mode data (LM) Whole-Body Acquisition List mode is an alternative way to store data during acquisition In the sinogram, events are counted in each matrix element. In List Mode, the two addresses (the detector numbers) of the two detectors involved are written to a continuous datastream. Every millisecond a timestamp is added to the stream Signals from cardiac or respiratory monitoring may be added Set of slices in multiple FOVs Step-by-step tabletop motion during acquisition Metastatic spread of cancer After the acquisition, the LM-file can be replayed and sorted into sinograms This allows dynamic and gated studies to be acquired without assumptions about time frames and number of gating bins In some cases it may even save raw data storage capacity $2010 Whole-body scan 55 Body Tomographic Planes Tomographic Planes for the Heart Frontal or Coronal Short Axis (SA) Transverse, Transaxial Transverse or Transaxial Sagittal Horizontal Long Axis (HLA) Frontal, Coronal Vertical Long Axis (VLA) Sagittal $2010 57 Comparison of $2010 Overview (4): PET and SPECT: ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| |||||||||||||||||||||||||||||||||||||| ||||||||||||||||||||||||||||||| PARALLEL HOLE COLLIMATOR + PET all directions simultaneously 1: Higher sensitivity, with ”electronic collimation” one direction at a time + SPECT 2: Fast dynamic scans 1: Dual isotopes possible 3: ”Biological” tracers with C11, N-13, O-15, F-18 2: Tc-99m generator + kits ! 58 Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) 3: Less expensive to run 4: Quantitative technique, with attenuation correction $2010 59 $2010 60 10 2D Imaging Michelogram – direct and cross slices The diagram shows all possible combinations of two detector rings. The ones ”allowed” are marked with a ”∗. Combinations that add to one sinogram are connected with a line between the points High Resolution Mode • Direct ∆Z = 0 • Cross ∆Z = ±1 • Limited Sensitivity Shown examples are the simple direct slices (diagonal in diagram) and High Sensitivity Mode • Direct ∆Z = 0, ±2 • Cross ∆Z = ±1, ±3 • Increased Sensitivity simple cross slices $2010 $2010 Direct Cross 62 2D / 3D Michelogram – 2D ”span 7” The number of combinations involved in direct and cross slices is known as the ”SPAN” -in this case 7 Note how the sensitivity decreases towards the edge of the axial Field-of-View $2010 63 2D ACQUISITION MODE septa employed low acceptance solid angle $2010 64 Michelogram – 3D, RD11 and RD15 2D/3D Volume Acquisition Detector ring 3D ACQUISITION MODE no septa High(er) acceptance solid angle Detector ring RD = Ring Difference In 3D, where larger angles (ring differences) are allowed, the michelogram becomes populated further away from the diagonal – eventually completely filled. $2010 2D acquisition (Septa installed) 3D acquisition (Septa removed) 65 $2010 66 11 Axial sensitivity profile ”Overlap” in whole-body scanning The overlapping scans add up to give a constant sensitivity and noise level (except for top and bottom) $2010 67 $2010 68 How to organize 3D data? 3D data – span 7 In 2D, data is a stack of sinograms. Event (X1,Z1,X2,Z2)--> Also for 3D, points in the michelogram may be joined, reducing the number of sinograms stored. z r Element r,θ of sinogram Z θ In 3D, data may be organized as a set of sinograms (Z1,Z2) or as a set of projection planes. Event (X1,Z1,X2,Z2)--> z v y θ φ u x Element u,v of plane θ , φ. $2010 69 $2010 u Projection Planes (u,v,θ,φ) or Z2 φ Z1 v u φ The image is no longer separable into a stack of 2D slices, but must be reconstructed as a volume. θ $2010 70 3-D PET: Reconstruction Problem 3D Data Organization Collection of sinograms (u,φ,Z1,Z2) Note: 3D u = 2D r, 3D v = 2D z, and 3D φ = 2D θ. 71 Problem: the cross-plane views are incomplete: they do not view entire image volume. $2010 72 12 3-D PET: The Reprojection Solution Forward Projected Data θ=0 A two-pass algorithm: θ= +/-1 θ= +/-2 θ= +/-3 θ= +/-4 Reconstruct in-plane data. Forward project to complete cross-plane views. 3D reconstruction of completed projection planes. (Kinahan and Rogers, IEEE TNS, 1988) θ= +/-5 θ= +/-6 As θ increases, the amount of forward projection required to complete data increases. θ = +/-5 chosen to balance data loss and amount of data completion. θ= +/-7 θ= +/-8 $2010 73 $2010 74 SSR = Single Slice Rebinning FORE = FOurier REbinning Uses the Frequency-Distance Principle, which states that the signal at (ω,k) in the 2D Fourier Transform of a sinogram arises primarily from activity at a distance r ∝ -k/ω from the center of the field of view. Simply assign data for each sinogram to the ”average position” – the transverse plane containing the intersection with the axial centerline. Good approximation near scanner center Large axial blurring towards the edge. Not acceptable for high resolution imaging. $2010 the center of the field of view. ω r y φ x 75 k 2D-FT Project $2010 3D-PET: The FORE Solution 76 FORE Algorithm Steps 1. Apply corrections 2. For each sinogram… – take 2D FFT. – rebin sinogram according to Frequency Distance Principle With r determined by the Frequency-Distance Principle, the axial displacement of the data on a cross-plane sinogram at oblique angle θ is calculated by ∆z = r tanθ. 3. Normalize and take 2D IFFT of 2D sinograms. z F FT sinogram sinogram image 4. 2D reconstruction of images θ $2010 r (x,y) 2D Recon 77 $2010 rebinned sinogram F -1 78 13 Overview (5): Image Reconstruction Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) Filtered Back-Projection methods: • 2D Filtered Back-Projection (2D FBP) • 3D Filtered Back-Projection (3D FBP) Iterative Reconstruction methods: • Maximum Likelihood Expectation Maximization (ML-EM) • Ordered Subsets Expectation Maximization (OSEM) • Full 3D iteration $2010 79 $2010 80 FAQ #1: Spatial Resolution Sources Ability to detect two objects close together. measured by phantoms What is the smallest [something] you can see in a PET scanner? Images Poor resolution Good resolution Line profiles of a point source Wrong question /No simple answer • Full Width at Half Maximum (FWHM) is the measure of resolution (unit: mm). Depends on the activity concentration/contrast • Typical PET resolution: FWHM ≈ 5 mm. FWHM 1/2 • Limitations are related to PET physics or scanner technology and design. $2010 81 $2010 82 Radial Vs Tangential Resolution Filtering is a balance between noise and resolution (FWHM) Radial resolution FWHM smaller ring diameter cause more degradation Tangential resolution Distance from center of FOV Plot of resolution versus radial position Smearing of point sources within the FOV 20 4 mm mm $2010 83 • Radial resolution degrades toward the edge of the field-of-view. • Tangential resolution is almost constant within the field-of-view. $2010 84 14 Radial Vs Tangential Resolution Results of modeling PSF in reconstruction (example: Siemens “High definition” recon) Crystal block Line of Response parallel to crystals axis Detector ring Line of Response oblique to crystals axis $2010 $2010 86 Mouse Heart PET/CT: 28g Mouse Sanchez-Crespo A et al. Image courtesy of David Stout Crump Institute for Molecular Imaging, Los Angeles CA Eur J Nucl Med Mol Imaging (2004) A similar device has been installed at the Panum Institute $2010 87 $2010 88 Overview (6): Scintillation detector Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) crystal γ light guide Dynodes with HV differences ∆V out (∆Ε) in photocathode 0 100 ns 200 γ - photon absorbed - converted to light – starts cascade af electrons $2010 89 $2010 90 15 Block Concept Evolution of PET-detectors Eγ X = A / (A + B) X1 X1 = 100 / (100 + 0) = 1.0 X2 X2 = 55 / (55 + 45) = 0.55 Therascan 4096+ 1981 (89) 2005 4*4 block 20*35 mm 13*13 block 6*12 mm total # 256 B A Biograph Hirez 1991 single detector 4*4 mm total # 4096 total # 24336 $2010 91 $2010 92 PET Detector Components Flood Histogram / Position Map Anger Logic Crystal elements and photomultiplier tubes A X= A+B Detector ring Y Detector block Y= C C+D Module array X $2010 Detector blocks and front-end electronics 93 PET Normalisation correction (example: GE Advance) – Compensates systematic efficiency variations (system geometry, crystal efficiency) PMT-gain = Detector module $2010 Energywindows 336 * Uniformity $2010 Every pair of blocks is tuned for coincidence timing Sinogram before normalisation correction 95 $2010 Sinogram after normalisation correction 96 16 Block contribution to sinogram All the contributions from a single detector will be on a line, angled as shown. Due to sensitivity differences a certain ”rhombic” structure will be visible in the sinogram. i cry sta l Siemens Biograph uses 68Ge cylinder One defective block will show up as a black, empty area in the sinogram i $2010 97 $2010 Effect of Normalisation 98 Uniform phantom – very sensitive to NORM without $2010 99 with $2010 Normalisation 100 Overview (7): (digression on the origin of ring artefacts) Red LOR’s connect block edges Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) (but first some theory of photon interactions with matter) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) Lines of Respons (LOR’s) $2010 101 $2010 102 17 Compton scattering Two processes stop photons in tissue: θ Simple, Absorption: e- γ Compton scattering: dθ γ’ no parameters Energy of outcoming photon can be calculated from the scattering angle Parameters e- γ E E’ and θ θ γ’ $2010 γ E’ (<E) 103 $2010 104 Scatter depends on: Klein-Nishina Cross sections Photon energy Next Slide Scatter angle θ Compton scattering is completely described by the Klein-Nishina cross sections: ! (α is incoming photon energy – in units of m0 = 511 keV) Wild physicists ahead ( 1 E ' (θ ) = E * 1 + α (1 − cos θ ) |||||||||||||||||||||||| 10 keV 100 keV ) 2 2 α 2 1 − cos θ d σ r02 1 = * 1 + cos 2 θ + d Ω 2 1 + α (1 − cos θ ) 1 + α (1 − cos θ ) 1MeV Physics: definition of attenuation absorbed no interaction Monoenergetic, parallel beam of photons scattered Attenuation = Absorption + Scatter $2010 107 $2010 108 18 Linear attenuation coefficient: µ Linear attenuation coefficient: µ I0 I(x) = I0* e-µµx - I0* e-µµx x absorbed x in cm, µ in cm-1 HVL = ln(2)/µ µ ∼ 0.7 / µ detected Water at 511 keV: µ ∼ 0.10 cm-1 1 cm ~ 10% scattered HVL ~ 7 cm Attenuation = Absorption + Scatter $2010 HVL is small compared to body dimensions µtotal = µabs + µscatter 109 $2010 110 Mass attenuation coefficient µm Mass attenuation coefficient µm (density: ρ g/cm3 ) The linear attenuation coefficient is proportional to density ρ. µm,,abs Dividing by ρ gives a number independent of density. and to µm for ice, fluid water, and steam has the same value To get proprotional to µm,,scatter: µ from a µm – curve or table: multiply by ρ $2010 111 1/E3 (energy) Z3 (atomic number) slowly decreasing with E independent of Z (1/E) $2010 112 Mass attenuation coefficients Trues Coincidence in water (soft tissue) and lead Scatter Coincidence Vary linearly with injected dose Tc-99m H2O Pb Tc-99m PET PET Not all counts are created equal... Random Coincidence $2010 Vary quadratically with injected dose there are the good, the bad, and the ugly counts 114 and liniearly with coincidence window 19 The ideal scintillation detector has High element no (Z) High density (ρ) Short decay of light High ligth output Low price Good absorption ⇒ crystal $2010 PET Detector Scintillator Materials light guide Crystal material High sensitivity ⇒ Less Deadtime, fast counting ⇒ Narrow coincidence window ⇒ less Randoms ⇒ Good energy resolution ⇒ Scatter rejection NaI:Tl BGO LSO:Ce LYSO GSO:Ce BaF2 max Z effective Z 53 83 71 71 64 55 51 73 66 54 59 54 density g/cm3 output decay time photons/keV ns 3.7 7.1 7.4 5.4 6.7 4.1 230 300 40 53 56 0.8 40 8 28 28 7.5 2 dynodes 115 photocathode $2010 Scintillator Light Output 116 Noise Equivalent Counts (NEC) When correcting for randoms and scatter, counts are subtracted But the noise (uncertaincy) is added! Correction The result is of less value than if the same #counts were measured directly NEC is the number of GOOD counts to which the result is equivalent. Measured $2010 Corrected NEC 117 $2010 118 119 $2010 120 Formula for NEC 2 NEC = [Trues * (1 − Scatterfraction)] [Trues + 2 f * Randoms] NB: ”Trues” in this formula contains scatter Scatterfraction: from 8% (2D) to 50% (3D) f: $2010 fraction of FOV, covered by object 20 NEC reflects the noise! $2010 121 $2010 122 Overview (8): Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) no interaction absorbed scattered |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| || Attenuation : Too few photons Scatter: too many, but misplaced photons $2010 123 $2010 PET is a quantitative technique 124 Attenuation in a homogenous medium (although most people don’t care these days...) without attenuation IF all the necessary corrections are applied: Deadtime Geometry, Normalization Randoms subtraction Attenuation and scatter Decay … $2010 with attenuation (or with perfect correction) C oun ts 12 8 Without attenuation Attenuation, µ=0.16cm -1 4 0 -8 -4 0 4 8 Position (cm) 125 $2010 126 21 Anthropomorphic phantom $2010 BEFORE Correction 128 P1 = exp(-µx) Attenuation Correction AFTER Correction (r,θ) µ x Sinogram Angle +90° L-x L(r,θ) $2010 Conclusion: Correction may cure the patients… 129 P2 = exp(-µ(L-x)) 0° Position -90° $2010 130 Probability of a coincidence: Probability of a coincidence: P(r,θ) = P1 * P2 P(r,θ) = P1 * P2 = exp(-µx) * exp(-µ(L-x)) = exp(-µL) [independent of x !] = exp(-µx) * exp(-µ(L-x)) = exp(-µL) [independent of x !] - and therefore easily measured with external (rotating) pin/point source - and corrected: each line of response multiplied by its own exp(µ µL) P1= exp(-µx) P1= exp(-µx) Sinogram Sinogram x Angle µ +90° Angle +90° µ L-x 0° 0° L(r,θ) L(r,θ) P2= exp(-µ(L-x)) $2010 -90° Position 131 P2= exp(-µ(L-x)) $2010 -90° Position 132 22 GE uses 68Ge Pin sources If µ is not a constant (and it IS NOT): x L− x 0 x P1 * P 2 = exp( − ∫ µ ( x)dx) * exp( − ∫ L µ ( x)dx) = exp(− ∫ µ ( x)dx) 0 still independent of the source position x P1 = exp( − ∫ µ ( x)dx) GE Advance and GE Discovery LS 0 Transmission Sinogram GE 4096 (by hand) (3 pins, automated complex robotic arm) +90° Angle µ 0° L(r,θ) GE Discovery ST etc. (1 pin in trunk) L $2010 -90° x Position 133 $2010 CT scans are µ-maps (in HU) Conversion of CT Numbers to µ They can be used for attenuation correction – essentially noise free But: µ(511 keV) ≠ µ(80 keV) , therefore a scaling is required Soft tissue and bone (and contrast) scale differently Not one simple scaling factor, but a lookup table for µ(HU; kVp) 100 Mass attenuation cm2/g Water, scatter Water, photo 10 134 Water, total Bone, scatter Bone, photo Bone, total For CT values < 100 , materials are assumed to have an energy dependence similar to water For CT values > 100, material is assumed to have an energy dependence similar to a mixture of bone and water 0.200 Attenuation at 511keV P 2 = exp( − ∫ µ ( x)dx) 0.150 0.100 Water/air Bone/Water 0.050 1 10 100 1000 0,1 The green line shows the effect of using water scaling for all materials 0.000 -1000 -500 0 500 1000 CT number measured at 140kVp 0,01 Energy in keV $2010 135 $2010 136 Scatter in Attenuation Correction Benefits • • • • Non-corrected PET image $2010 SPECT Provides PET images closer to real radiotracer distribution Corrects attenuation-introduced geometric artifacts Improves lesion detection in deep organs Allows absolute quantitative PET studies appears ”inside” the object PET may appear outside object Attenuation-corrected PET image Better Image Quality $2010 138 23 Scatter korrektions teknikker Scatter Correction by Function Fitting Metz / Wiener filtering Asymmetrical energy window Dual energy window Dual photopeak window Weighing of detected events (WAM) Channel ratio method Photopeak energy distribution analysis Position dependent scatter correction Stationary / Non-stationary deconvolution Iterative reconstruction techniques Neural network Pixel by pixel spectral analysis Regularized deconvolution-fitting method Scatter-free imaging (CFI) Holospectral imaging Factor analysis of medical image sequences Fit data to projection tails This requires tails! More elaborate methods model the scatter from detailed knowledge of the object $2010 139 $2010 140 Scatter: The Problem in 3D Overview (9): 2D: 20 kcps (18 true, 2 scatter) 3D: 200 kcps (135 true, 65 scatter) 3D: 26 kcps (0 true, 26 scatter) Reconstruction from projections Positron imaging history Sinograms – various types of acquisitions 2D/3D, FBP vs. iterative recon methods, image presentation Resolution (how small ... can you see) Scanner physics (how to optimize a PET scanner) Good counts, bad counts (!) and noise equivalent counts (NEC) Attenuation and scatter correction of PET-data PET/CT scanners (PET/MR, and SPECT/CT) Sources outside FOV very significant. Deconvolution or modeling methods cannot see this activity. $2010 141 Good reasons to combine $2010 142 Functional Images PET and CT: PET Images 0: 1: 2: 3: “abnormal activity” Because it is possible (MR more difficult) Improved attenuation correction in PET Complementary information structure + function anatomy + physiologi/biochemistry Exact localization (”free” coregistration) Weather Patterns “weather activity” $2010 143 $2010 144 24 Structural Images Fusion Images CT Images Discovery PET/CT Images “precise body’s anatomy” “abnormal activity and precise body’s anatomy” Geographic Map USA Weather Map “precise outlines of the states” “intense weather and precise outlines of the states” $2010 145 $2010 PET/CT scanner: 146 PET/CT (or SPECT/CT) scanner: 168 cm First “physicist’s” PET-CT was constructed by Townsend, Beyer, Kinahan et al in Pittsburgh 1994- 200 cm We (Rigshospitalet) got a prototype from GE in 2001 Today commercial devices exist from: 182 cm (Townsend et al 2004) GE Siemens-CTI Philips and a few others - with more than 2000 installations worldwide Anatomical localisation of tracer – Attenuation correction Since 2005, almost NO stand-alone PETs have been installed $2010 combines two modalities in one gantry, common axis, common bed allows precise fusion of images provides structure + function in the same image can use CT for attenuation correction of PET /SPECT) 147 $2010 PET/CT: GE Discovery LS 148 Inside Discovery LS Advance NXI + Lightspeed plus Detector material BGO $2010 149 $2010 150 25 PET/CT PET –CT alignment /registration RH (4) Bispebjerg (2) Næstved Herlev (2) Not likely to change spontaneously. Århus PET (2) Hillerød Århus NUK Gentofte Herning? Vejle (2) For GE and Siemens: Only opened by service. Must be measured after ”invasive procedures” Glostrup? Philips Gemini (KAS Herlev) (Allegro + MX8000 D) Siemens Biograph 16 (RH) Odense (3) Philips Gemini designed to scan in the separated position also. Ålborg Køge (Hamlet) Århus Skejby Hvidovre? $2010 GE Discovery ST GE (Milano San Raffaele) 151 $2010 152 Siemens PET – CT alignment PET – CT alignment GE Discovery aligns CT with PET transmission using the built-in Ge-68 line sources. Siemens Biographs have no transmission sources. No (other) activity is required The phantom contains 5 glass spheres Instead 2 Ge-68 line sources mounted in a black box are used. Newest GE-scanners use similar phantom with Ge-68 spheres $2010 153 They are visible on both PET and CT, and an automated algorithm provides the correction $2010 Philips PET – CT alignment 154 Small Animal PET-CT Special service tool holding 6 Na-22 point sources Registration performed at 1 year interval CT resolution down to 15 µ $2010 155 $2010 PET resolution < 2 mm 156 26 Potential PET-MR configurations PET-MR: the ”cheap” solution Based on: Eur J Nucl Med Mol Imag (2009) Suppl 1 S86 $2010 157 $2010 (real) PET-MR is coming 158 (real) PET-MR is coming Siemens has a prototype combination: 3 T MR scanner with brain coil PET ”insert” between coil and magnet Requires detector insensitive to magnetic field Attenuation correction is a challenge: MR does not describe density, and coil is not measured Whole Body model is now being tested $2010 Presented at EANM 2008 by B. Pichler 159 $2010 160 DOI measurement with APD PSAPD (Depth of Interaction – to avoid parallax-errors) Position Sensitive Avalanche Photo Diode Presented at EANM 2008 by A. Del Guerra From: Dokhale et al. Phys Med Biol 49 (2004); 4293-4304 $2010 161 $2010 162 27 Basic PET Detector SSPMs based Solid-state photo-multipliers (SSPMs) can be fabricated from small silicon sub-pixels to replace PMTs making them attractive for PET+MR and TOF-PET: • fast, low-jitter time response • magnetic field immunity • small form-factor Coincidence timing res. • Readout circuits/ASIC development Magnetic Sensitivity • Multiplexing options Noise • Handling multi-crystal events Technology Maturity • MR compatible architecture 25% 50% 50% 550ps 350ps 1000ps 106 105-106 102 100 mm 2 mm 2 mm High Low Low Gain Profile Technical challenges: SSPM Detection efficiency Next years fashion? APD PMT Low Low High 40 yrs 4 yrs 8 yrs Low High High From: Cost MIC2009 $2010 SPECT/CT (GE Hawkeye) SPECT/CT 164 X-Ray Tube is mounted on the same gantry as the gamma cameras serves the same dual purpose of A ‘CT scan’ takes 15 s / cm With 1 inch crystals, “PET” is possible Anatomical localisation of tracer and Attenuation correction AC is not as simple or ”exact” as in PET - because we have only one photon $2010 165 $2010 BrightView XCT SPECT/CT SPECT/CT Siemens Symbia 166 Philips Precedence = Skylight SPECT + Brillance CT $2010 167 $2010 168 28 PET / SPECT / CT PET / SPECT / CT for small animals Mediso Budapest Exhibition at EANM 2008 $2010 169 $2010 170 29