William Hallet - PRIMA IV 1

Transcription

1 Quantitative and application specific imaging PET: the measurement process,reconstruction, calibration, quantification Dr William Hallett Centre for Imaging Sciences Imperial College Hammersmith Hospital Radiochemistry + GMP PET-CT + MR + GCP Preclinical PET-CT Integrated IT system 60 Staff Mathematics Physics IT Chemistry Biology Clinical Imaging services for academic and industrial partners William Hallet - PRIMA IV 1

2 Justification for medical exposures Risk from radiation dose must be balanced against benefit to individual and/or society Medical radiation exposures need to be justified and optimised Staff must be trained and qualified Approval for activities required Ethical approval for research studies Overview of the whole process Input function Quality Control Scanning Chemistry Cyclotron IT system Clinical area Image Analysis Blood Labs Recruiting and Screening William Hallet - PRIMA IV 2

3 Isotopes commonly used in PET half-life (m) source Some PET applications 18 F cyclotron Glucose metabolism brain metabolism, tumour imaging, inflammation (FDG), tumour proliferation (FLT), bone imaging (F - ), hypoxia (F-miso), Neuroreceptor imaging, AD, peptides, gene expression (reporter gene imaging) 11 C 20.4 cyclotron Organic compounds in general, tumour proliferation and metabolism (thymidine, methionine, choline), AD (PIB),neuroreceptors 15 O 2.1 cyclotron Blood flow (water, CO 2 ), blood volume (CO) 13 N 10 cyclotron Myocardial blood flow (NH 3 ) Less familiar positron emitting isotopes half-life (m) source Some PET applications 68 Ga 68.3 generator Lung permeability, tumour imaging (hormone analogues) 66 Ga 567 cyclotron 82 Rb 1.25 generator Tumour imaging (citrate, hormone analogues), antibodies, proteins, BBB integrity Myocardial blood flow 94m Tc 52 cyclotron 62 Cu 9.76 generator Nuclear medicine procedures instead of Tc 99m eg myocardial perfusion, antibodies Hypoxia eg tumours (ATSM etc) 64 Cu 768 cyclotron Antibodies, proteins, peptides, hypoxia (ATSM etc) 124 I 6000 cyclotron Tumour imaging (amino acids), antibodies 76 Br 966 cyclotron Neuroreceptors, antibodies William Hallet - PRIMA IV 3

4 Positron Emission Fundamental process nuclear decay by positron emission. Unstable proton rich parent nucleus Proton decays to neutron N e+ Positron and neutrino emitted Positron annihilates with electron e- energy and momentum conserved photons have equal energy electron rest mass 511KeV Positrons travel ~1mm in tissue before annihilating localisation PET it s particle physics! Proton inside unstable parent nucleus quarks uud 1: Strong force holds quarks together W + e+ 3: Electromagnetic force turns positron and electron into two photons e- 2: Weak force changes u quark into d quark Neutron Neutrino quarks udd William Hallet - PRIMA IV 4

5 A large particle physics experiment looking for Higgs bosons The Large Hadron Collider at CERN The Atlas detector at the LHC A small particle physics experiment Looking for positrons A clinical PET-CT scanner PET is very sensitive ~ 3% chance each nuclear event is detected correctly but millions of events are needed to get high quality images William Hallet - PRIMA IV 5

6 PET gamma rays are energetic and penetrating. protection of staff from radiation required Positron Emission Tomography Gamma rays interact with detectors - scintillator crystals Thousands of detectors arranged around patient (eg 4 rings, detectors) Count millions of coincidences to determine radionuclide distribution William Hallet - PRIMA IV 6

7 Development of PET 1930 Cyclotron Laurence et al 1953 Annihilation coincidence detection Brownell + Sweet 1966 Regional localisation Yamamoto et al 1975 Transaxial tomography Ter-Pogossian, Phelps, Hoffman C deoxyglucose Sokoloff et al FDG PET Reivich et al 1995 PET-CT Townsend et al First commercial PET scanners William Hallet - PRIMA IV 7

8 Rapid technological development - PET-CT Siemens Philips GE Preclinical PET-CT system from Siemens Inveon PET system with multi-modality CT system William Hallet - PRIMA IV 8

9 Block detector building block of a PET scanner Used in the majority of installed PET systems Cost effective and practical solution to reading out scintillation light from a very large number of small crystal elements with a small number of photo-multiplier tubes Gamma ray interacts with scintillator crystal ( eg BGO, LSO, LYSO, GSO, GYSO etc) Light detected by small number of PM tubes able to localise gamma ray event t f th t l l t Primary interaction of gamma rays with matter - Compton scatter Gamma ray energy >> electron binding energy Scattered photon E< 511 KeV Scattered electron : energy absorbed in crystal or patient e- Electron e.g. in crystal or subject Incident 511 KeV photon Scatter can occur in the detector or the subject being imaged William Hallet - PRIMA IV 9

10 PET singles data energy spectrum Counts Backscatter peak (170 kev) Energy Window Photopeak (511 kev) Compton edge (341 kev) FWHM Energy resolution = FWHM/Photopeak Compton edge maximum energy lost by photon scattered through 180 Energy Need to include scattered events (many occur in the detector) Coincidence detection PET scanner looks for coincidence events pairs of gamma rays that strike different detectors within ~ 10-9 seconds of each other Has to be able to do this ~ 100,000 times a second Single coincidence circuit between individual detectors PET scanner has many coincidence circuits William Hallet - PRIMA IV 10

11 Sensitivity PET has higher sensitivity than single gamma ray imaging eg SPECT Collimator Gamma camera - Single photon detector PET scanner coincident gamma detector electronic collimation PET data is organised into parallel lines of response Angle around scanner Parallel lines of response at different distances from the centre of the scanner William Hallet - PRIMA IV 11

12 PET data can be stored as event list or Sinogram 01xxx 00xxx 01xxx 01xxx 00xxx Listmode each detection event gives LOR regular timing events (clock) other information eg gate signals Each location in the Sinogram corresponds to a LOR The Sinogram is the Radon Transform of the object True, Scattered, and Random events Scatter True Random Line of response (LOR) William Hallet - PRIMA IV 12

13 Prompt, Delayed and Trues Sinogram data Sinograms of a point source Prompt events = Trues + Randoms + Scatter Delayed events = Random non-coincident events Trues = Prompts - Delayed - Scatter Angle Distance Scatter Scatter fraction: Fraction of prompt events within the energy window that are scattered (recorded in the wrong place) True counts Scattered counts measurement of scatter fraction on a preclinical scanner using a line source in a scatter phantom William Hallet - PRIMA IV 13

14 Scatter and randoms degrade PET scanner performance They add statistical noise that cannot be removed Noise Equivalent Count Rate Count rate curves for 3 ring (dashed lines) and 4 ring (solid lines) Scanners Trues 2 Trues+2Randoms+Scatter NECR (signal/noise) 2 Randoms Activity 2 Scatter fraction ~ constant At high activities deadtime important measured according to NEMA NU Scanner performance depends on the object Ideally should be able to tune scanner settings to the subject being scanned Rat phantom Mouse phantom NECR curves for a preclinical PET scanner with different sized phantoms and different energy and timing windows from Tai et al JNM 2005; 46:455 William Hallet - PRIMA IV 14

15 2D and 3D acquisition 3D acquisition Current generation of scanners using LSO etc No septa High sensitivity (x10) High randoms & scatter - use higher timing & energy resolution Scatter fraction can still be 30-40% in the body Susceptible to out of field of view activity - relies on scatter modelling Harder to reconstruct the image - solved 2D acquisition Previous generation of scanners using BGO scintillator Inter-plane lead septa stop oblique gamma rays Low sensitivity Low randoms & scatter - not so reliant on corrections 3D scanner sensitivity - increases with axial field of view 4 ring 3 ring PET scanner axial sensitivity profiles for a 3 ring and 4 ring system William Hallet - PRIMA IV 15

16 Attenuation Half of 511KeV gamma rays are attenuated every 7cm in tissue Both gamma rays must be detected to register a coincidence Attenuation leads to reduction in apparent activity in the centre of large objects Attenuation factor across the chest may be as high as 50 Many attenuated events in PET are scattered elsewhere In PET (unlike SPECT) the attenuation correction factors can be measured directly using an external source of radiation Attenuation correction: needed for quantification Annihilation photons from PET are absorbed within patient Correction for this is needed to get representative image Non-corrected Non-corrected Attenuation corrected Attenuation corrected William Hallet - PRIMA IV 16

17 Attenuation correction methods Use 68 Ge (positron emitting) source transmission scan + same energy as PET image - slow, noisy image single photon source (e.g. Cs137) can also be used Use CT images (x-ray transmission scan) + Rapid, gives extra clinical and positional information - Energy is ~70 kv, c.f. 511 kv PET image, higher dose CT data CT measures attenuation properties of the medium - depends on electron density Images: visible human project mm William Hallet - PRIMA IV 17

18 Inside the gantry x-ray tube HV generator x-ray detectors PET-CT uses CT scanner to correct for PET attenuation Tube Aperture Detectors William Hallet - PRIMA IV 18

19 CT X-ray detector x-ray Visible light Electrical signal output detector attenuation detector William Hallet - PRIMA IV 19

20 CT data acquisition CT measures the attenuation of X-rays (not gamma rays) through the body Projections through 180/360 form a sinogram Multi-slice helical CT as bed feeds through scan planes trace out helical path in the subject William Hallet - PRIMA IV 20

21 Image reconstruction: back projection Reverse of the measurement process needs many events Original object 2 projections 4 projections 8 projections 16 projections 32 projections Note blurring : extended point spread function (PSF) Filtered back projection Filtering of projection data to counteract blurring : filter is inverse of the PSF 2 projections 4 projections 8 projections Filter 16 projections 64 projections 64 projections (not filtered) William Hallet - PRIMA IV 1

22 CT today Courtesy: Toshiba Courtesy: Siemens CT Scan Speed Improved scan speed Better coverage in a single breath hold Reduction in patient movement artefacts Better use of contrast media Less need for sedation in paediatrics e.g. Aortic dissection: 500 mm in 16 sec 4 x 2.5mm slices, pitch Courtesy: Dr. Baum, University of Erlangen William Hallet - PRIMA IV 2

23 CT Z-axis resolution Slice widths of mm available for routine scanning Isotropic imaging now possible Simulated effect of CT dose reduction Scanned dose: 1 Simulated dose: Images courtesy Y. Muramatsu, NCC Tokyo William Hallet - PRIMA IV 3

24 Converting CT to PET attenuation maps Non-linear relationship between CT number and 511 kv attenuation Segmented vs non-segmented attenuation maps CT attenuation correction for PET CT blurred to match the final PET image resolution Segmented into regions with different attenuation coefficients (via a lookup table) Resulting attenuation map forward projected and multiplied with the PET sinogram William Hallet - PRIMA IV 4

25 CT also gives anatomical localisation information for PET Modern PET-CT scanners can detect small lesions in large patients! 305 lb Patient "Courtesy of The University of Texas, M.D. Anderson Cancer Center, PET Facility, Division of Diagnostic Imaging" William Hallet - PRIMA IV 5

26 Preclinical PET-CT Images given honourable mention in Siemens preclinical image of the year competition 2010 submitted by UT Southwestern medical Center What does PET measure? PET measures radioactivity concentration eg Bq/ml with time (1s or more) PET images are usually corrected for radioactivity decay (e.g. to injection time) Semi-quantitative measurements: Standardised Uptake Values (SUV) activity concentration/injected activity per gram no measurement of tissue input function hence approximate 1 for a uniform phantom mostly used for 18 F-FDG oncology imaging variants exist that try to account for physiological effects Analysis of dynamic PET data (4D) enables absolute measurement of biological processes in vivo See next talk William Hallet - PRIMA IV 6

27 Reconstruction by back-projection Projections are the original image summed along each LOR in a particular direction Back-project duplicate value of projection along each LOR Doing this for all projections (angles) gives an approximate reconstruction Projection 2D Point Spread Function Consider a point source Back projection gives a blurred image PSF = density of back-projected rays ~ 1/r in 2D Scanner transaxial plane LOR 1 r LOR 2 LOR 3 William Hallet - PRIMA IV 7

28 Central Slice Theorem FT of projection (P) = central slice through FT of the object (I) True in any number of dimensions eg 2D and 3D p k x Object I O P FT(I) Projections and Sinograms z FOV Detector ring B Projection slice Sinogram slice Sinogram - fixed ring pair Detector ring A For small Field Of View and oblique angle projections and sinograms are orthogonal slices through the same 4D data set William Hallet - PRIMA IV 8

29 D point spread function PSF=density of back-projected rays ~ 1/r in 2D ~ 1/ r 2 in 3D Scanner transaxial plane LOR 1 r LOR 2 Scanner axis LOR 3 3D PSF = 0 outside shaded area Redundancy of 3D projection data p is unique p is not unique k O p k O p } 2D 3D William Hallet - PRIMA IV 9

30 Truncation of 3D projection data In practice only the direct ( = 0 projections are complete Missing data LOR Missing data Rebinning If 3D oblique projection data can be rebinned into the direct (oblique angle 0) projections a 2D reconstruction algorithm can be used This can be achieved using the Central Slice Theorem Data at spatial frequency vector k can be found from both oblique and direct projections k FT(I) O FT of oblique projection William Hallet - PRIMA IV 10

31 Resolution a steep hill to climb distance travelled by positrons in tissue before annihilating (mm or so - fundamental photon accolinearity due to residual momentum of electron - positron pair (fundamental) depth of interaction - uncertainty in position of interaction of gamma rays with thick crystals pile-up in detector electronics leading to uncertainty in positioning scattering between detector elements size of detector elements image signal/noise ~(detector element width) 2 x dose For PET no point making detectors 1mm Image resolution is typically worse than this... Sinogram profile of point source in clinical PET scanner Preclinical system resolution F-18 in tip of glass capillary tube Resolution William Hallett - PRIMA IV 1

32 Resolution the Partial Volume Effect Reduction in apparent activity for objects less than twice the image resolution Statistical Reconstruction Iterative methods attempt to maximise the statistical likelihood of the image 2 Iterations, 8 Subsets 8 Iterations, 8 Subsets 8 Iterations, 21 Subsets 1. Take a reasonable guess at the image 2. Assuming a distribution of events, eg Poisson, calculate how likely the data (sinogram) is given this image 3. Calculate a new image that maximises this likelihood 4. Repeat speed up by using only some of the data each time OSEM (ordered subsets expectation maximisation) Resolution William Hallett - PRIMA IV 2

33 Point Spread Function Reconstruction - Resolution Recovery Iterative methods can incorporate more sophisticated models of the imaging process with PSF modelling in reconstruction Standard iterative reconstruction For a point source in the scanner the scanner response changes with position ie is non-stationary Building this non-stationary scanner response into the reconstruction improves the resolution significantly (> factor of 2) Images provided by Siemens Latest PET scanners have astonishing resolution (with PSF modelling) ~2mm resolution depends on accurate statistical modelling of the data acquisition process Text Resolution William Hallett - PRIMA IV 3

34 MAP reconstructions p(a) prior probability of A p(b) prior probability of B p(a B) probability of A and B = p(b A) probability of B and A conditional probabilities p(b A) probability of B given A posterior probability of B given A p(a B) probability of A given B posterior probability of A given B p(b A) = p(b A) x p(a) p(a B) = p(a B) x p(b) p(b A) = p(b) x p(a B)/p(A) Bayes Theorem p(image data) = p(image) x p(data image) / p(data) maximum likelihood of the data given the image Prior probability of the image suppress noise by making smooth images more likely - regularization/penalisation make images correspond to anatomy eg by weighting in favour of images that resemble CT or MR MAP Reconstructions FBP MAP Images courtesy of Metropolis Data Consultants Resolution William Hallett - PRIMA IV 4

35 Evaluation of different reconstruction algorithms Clinical Image Comparison Back-projection OSEM 8i,8s OSEM+PSF 8i,21s Evaluation of different reconstruction algorithms Based on phantom images Hot Sphere Contrast (HSC) Measured Activity True Activity Cold Sphere Contrast (CSC) 1 Measured Activity True Background Activity System Linearity HSC CSC Background Variability SD (60 Background Regions) Mean Background Resolution William Hallett - PRIMA IV 5

36 Effect of reducing administered activity Sample Listmode data to mimic effect of reducing the injected dose Figure 1: Simulation Process Derived kinetic parameters can be remarkably stable as dose reduced Uncertainty in fitted parameters increases as expected Motion correction - gating period motion (breathing, cardiac cycle) bin data into frames that correspond to the same anatomical position end-systole end-diastole ungated Typically requires an exterior signal or measure of position - such as an ECG trace Resolution William Hallett - PRIMA IV 6

37 Motion leads to artefacts in parametric imaging Herzog et al JNM :1059 PET calibration Administered Activity Tissue Activity Arterial/venous Blood Activity NB typically ~200MBq given Blood Sample Activity PET measurements are only reliable if these all agree with each other Resolution William Hallett - PRIMA IV 7

38 Things can go wrong... Daily QC Ge-68 phantom Scanner correctly setup Scanner calibration factor tries to compensate Scanner incorrectly setup Time of Flight PET x D Annihilation photons originating away from the centre of the scanner arrive at different times t=2 x/c Timing resolution of current generation of PET scanners is ~500 picoseconds speed of light 3 x 10 8 m/s Can localise events to ~8cm Effective gain in counts (sensitivity) is diameter of object/ x Resolution William Hallett - PRIMA IV 8

39 Time of Flight PET reconstruction without TOF information With perfect timing resolution (no blurring) no reconstruction would be necessary! TOF data also includes information about attenuation with TOF information Figures courtesy of Siemens PET-MR - Simultaneous multi-modality data acquisition High resolution artefact free PET images High resolution artefact free MR images Commercial clinical and preclinical PET-MR systems are now available Images from Siemens prototype head system 2.0mm 1.5mm 2.5mm 1.0mm 3.0mm 3.5mm MR PET Resolution William Hallett - PRIMA IV 9

40 Acknowledgements: Nick Keat of Imanova for CT slides and other contributions Resolution William Hallett - PRIMA IV 10