Question: Describe the basic principles of PET imaging.
Answer:
Positron Emission Tomography (PET) is a functional nuclear medicine imaging modality that
uses a uniquely quantum-mechanical event, the annihilation of a positron with an electron to
produce pairs of collinear 511 keV photons. These photon pairs are detected in coincidence by a
ring of radiation detectors encircling the patient, generating lines of response (LORs) from which
a three-dimensional map of radiotracer distribution is reconstructed. PET stands apart from
planar scintigraphy and SPECT because it uses electronic collimation rather than physical lead
collimation, yielding higher sensitivity and intrinsically superior spatial resolution.
The clinical dominance of PET in oncology, neurology, and cardiology rests on this physical
foundation. The single most important radiopharmaceutical remains ¹⁸F-FDG (¹⁸F-
fluorodeoxyglucose), but the same principles apply to every PET tracer: ¹⁸F-PSMA, ¹⁸F-NaF,
Ga-68–DOTATATE, Cu-64–ATSM, and the rapidly expanding theranostic agents.
Positron Emission: The Nuclear Physics Basis
PET tracers are labelled with proton-rich, neutron-deficient radionuclides that undergo β⁺ decay
(beta-plus decay). In this mode of radioactive transformation, a proton within the nucleus
converts to a neutron with the simultaneous emission of a positron (e⁺) and an electron neutrino
(ν). The nuclear equation is:
p⁺ → n + e⁺ + νₑ
For β⁺ decay to occur spontaneously, the parent nuclide must be energetically higher than the
daughter by at least 2 × 0.511 MeV = 1.022 MeV (the rest-mass energy of the positron–electron
pair). This explains why β⁺ emitters have relatively short half-lives and are produced by
cyclotron bombardment or generator systems. The kinetic energy given to the positron ranges
from zero up to the Q-value minus 1.022 MeV, and the emitted positrons have a continuous
energy spectrum up to this maximum.
Key PET radionuclides and their properties are: ¹⁸F (T½ = 109.8 min, E₋₊ⁿₑₜ = 633.5 keV, β⁺
abundance 97%); ¹¹C (T½ = 20.4 min, E₋₊ⁿₑₜ = 960 keV); ¹³N (T½ = 9.97 min); ¹⁵O (T½ = 2.04
min); Ga-68 (T½ = 67.7 min, generator-produced from Ge-68); Rb-82 (T½ = 76 sec, cardiac
perfusion); Cu-64 (T½ = 12.7 hr, theranostics).
Positron Range and Annihilation
The emitted positron traverses tissue, losing kinetic energy through ionisation and excitation
interactions with atomic electrons (the same mechanisms as any charged particle). As it slows, it
follows a tortuous path and eventually reaches thermal energies, at which point it undergoes
annihilation with a nearby electron. The mean range in tissue before annihilation is determined
by the maximum positron energy: for ¹⁸F it is approximately 0.6 mm (FWHM of positron range
blur), for Ga-68 approximately 2.9 mm, and for Rb-82 approximately 5.9 mm. This range
constitutes a fundamental, irreducible limit on PET spatial resolution, which is why ¹⁸F-labelled
tracers consistently produce sharper images than Ga-68 or Rb-82.
At annihilation, the combined rest-mass energy of the positron–electron pair (2 × 511 keV =
1022 keV) is converted into two 511 keV annihilation photons emitted in opposite directions
(180° ± 0.25°) to conserve momentum. The slight angular deviation from perfect collinearity
arises because the positron and electron are not completely at rest at the moment of annihilation;
they retain small residual momenta. This non-collinearity blurs the reconstructed image by
approximately 1–2 mm at the radius of modern clinical scanners (typically 40–50 cm detector
diameter). Energy conservation demands each photon carry exactly 511 keV, which is the basis
of the energy window used for detection.
Figure 1. Positron Emission, Range, and Annihilation Producing Collinear 511 keV Photon Pairs
The proton-rich nucleus emits a positron that travels 0.6–6 mm before annihilating with an electron, producing two
511 keV photons separated by approximately 180°, which are detected in coincidence by the surrounding detector
ring.
Examination Significance: Frequently drawn to introduce PET physics. Examiners expect labelled depiction of
positron range, the annihilation event, collinear photon pairs, and the line of response (LOR). Non-collinearity as a
resolution-degrading factor is a common viva question.
Coincidence Detection and the Line of Response
The defining feature of PET is electronic collimation. Rather than using a lead septa to define
the direction of accepted photons (as in SPECT), a PET scanner accepts only those photon pairs
that trigger two opposing detectors within a narrow coincidence timing window (typically 6–12
ns). If two detectors respond simultaneously within this window, the scanner draws an imaginary
line — the line of response (LOR) — between them and infers that an annihilation occurred
somewhere along that line. The collection of millions of LORs over the acquisition period forms
the raw dataset from which tomographic images are reconstructed.
This approach has a profound practical consequence: because sensitivity scales with detector
solid angle rather than lead collimator geometry, PET achieves a sensitivity approximately 20–
50× higher than SPECT for the same injected activity. The trade-off is that any photon pair
arriving within the coincidence window may be accepted, making PET susceptible to three
categories of coincidence event that must be corrected in post-processing.
True, Scattered, and Random Coincidences
True coincidences occur when both 511 keV photons from a single annihilation event reach
opposing detectors within the coincidence window without prior scatter. True coincidences carry
valid spatial information and define the true count rate (T).
Scattered coincidences occur when one or both photons undergo Compton scatter within the
patient before reaching the detector. The photon is deflected from its original trajectory, so the
LOR drawn by the scanner is incorrect. Scatter does not change the photon energy enough to fall
outside the energy window (especially for soft-tissue scatter), so it represents a source of
background noise and image blurring. The scatter fraction is typically 30–40% in whole-body
PET/CT and is corrected by model-based scatter simulation or tail-fitting methods.
Random (accidental) coincidences arise when two photons from different, unrelated
annihilations happen to reach opposing detectors within the same coincidence window.
Randoms add a spatially uniform background that degrades contrast, especially at high count
rates. The random coincidence rate (R) is given by:
R = 2τ × N₁ × N₂
where τ is the coincidence time window (in seconds), and N₁, N₂ are the single-detector count
rates in detector pairs 1 and 2. Since randoms scale with the square of count rate (through N₁ ×
N₂), they become disproportionately problematic at high injected activities. The standard
correction is the delayed coincidence channel method: a second coincidence processor uses a
delayed copy of one detector signal (delayed by ~100 ns, well outside the true coincidence
window), counting only randoms. This randoms estimate is then subtracted from the prompt
coincidence data.
Figure 2. True, Scattered, and Random Coincidences in PET
True coincidences yield a correct LOR; scattered coincidences produce a misassigned LOR due to photon
deflection; random coincidences from two independent annihilations generate a spurious LOR that adds uniform
background noise.
Examination Significance: This three-way classification is tested at every postgraduate examination. Candidates
must define each type, state its effect on image quality, and describe the correction method. Scatter fraction values
and the random rate formula are specific examination targets.
The PET Detector System
Because 511 keV photons are substantially more energetic than the 140 keV gamma photons
detected in SPECT, PET demands dense, high-atomic-number (high-Z) scintillator crystals with
fast decay times. The three requirements in tension are: stopping power (determined by density
and Z), timing resolution (determined by light output and scintillation decay time, which
governs time-of-flight capability), and energy resolution (governs the ability to reject scattered
photons with the energy window).
Scintillator Crystals
The original PET scintillator, BGO (bismuth germanate, Bi₄Ge₃O₁₂), offered excellent
stopping power (density 7.13 g/cm³, effective Z = 74) but very slow scintillation decay (300 ns),
precluding time-of-flight (TOF) capability. LSO (lutetium oxyorthosilicate, Lu₂SiO₅:Ce) and
its close analogue LYSO (lutetium–yttrium oxyorthosilicate) revolutionised PET by
combining high density (7.4 g/cm³), excellent stopping power, very high light output, and a fast
decay constant of 40 ns. This combination makes LYSO the de facto standard for modern
clinical TOF-PET/CT. GSO (gadolinium oxyorthosilicate) and LaBr₃ (lanthanum bromide)
offer superior energy resolution and are used in research applications.
Each detector block typically consists of a matrix of small crystal elements (e.g., \u00B44 × 4
mm² for clinical systems) coupled to photomultiplier tubes (PMTs) or, increasingly, silicon
photomultipliers (SiPMs). SiPMs are semiconductor photon detectors with internal gain
produced by Geiger-mode avalanche breakdown; they are inherently MR-compatible (no
photocathode, no vacuum envelope, insensitive to magnetic fields) and have enabled the design
of integrated PET/MR systems. SiPMs also achieve faster coincidence timing (~300 ps),
extending the value of TOF information.
Energy Window and Pulse Height Analysis
Each detected scintillation is processed through a chain of preamplifier → shaping amplifier
→ pulse height analyser (PHA). The PHA applies an energy window centred on 511 keV,
typically 350–650 keV (i.e., ±27%) for BGO-based systems and 425–650 keV for LYSO-based
systems. This window rejects scattered photons, which have lost energy through Compton
interactions, and also rejects low-energy electronic noise. Photons that pass the energy window
are called singles; coincident singles in two opposing detectors within the timing window form a
valid coincidence event.
Figure 3. PET Scanner System Block Diagram: Detector Signal Chain to Image Reconstruction
Annihilation photons are detected by scintillator–photodetector pairs, amplified and pulse-height analysed, then
assessed for coincidence; valid LOR data are corrected and reconstructed into tomographic images by iterative
algorithms such as OSEM.
Examination Significance: Examiners routinely ask candidates to draw or describe the PET signal chain. Key
components to label are: crystal, PMT/SiPM, preamplifier, CFD (constant fraction discriminator), PHA,
coincidence processor, and reconstruction. TOF capability belongs at the coincidence processor stage.
Time-of-Flight (TOF) PET
In conventional PET, once a valid LOR is established, the scanner assumes the annihilation
occurred somewhere along the chord—information about exactly where is lost. Time-of-flight
PET exploits the small timing difference (Δt) between the arrival of the two 511 keV photons at
opposing detectors to localise the annihilation along the LOR. The localisation uncertainty is
given by:
Δx = (c × Δt) / 2
where c is the speed of light (30 cm/ns) and Δt is the coincidence timing resolution (CTR). A
CTR of 200 ps corresponds to a spatial localisation uncertainty of:
Δx = (30 cm/ns × 0.2 ns) / 2 = 3 cm
Rather than back-projecting uniformly along the entire LOR (as conventional PET does), TOF-
PET places most of the activity estimate within a 3 cm Gaussian kernel centred on the most
probable annihilation point. For a 40 cm diameter patient, conventional PET uses the entire 40
cm LOR while TOF-PET confines reconstruction to a 3 cm segment—an SNR gain factor of
approximately D/Δx √(D/Δx) = √(D/Δx) ≍ √(40/3) ≈ 3.65× for a 200 ps system. In clinical
practice this translates to a 2–3× improvement in SNR for large patients, allowing dose
reduction, faster scans, or improved lesion detectability in obese patients. Modern LYSO/SiPM
scanners achieve CTR of 210–250 ps; research systems with digitalSiPM or PbWO₄ crystals are
approaching 100 ps.
Data Acquisition Modes
2D mode (historical): Lead or tungsten inter-plane septa were inserted between detector rings to
reduce out-of-plane scatter. This reduced sensitivity but improved scatter fraction. Virtually no
modern system operates in 2D mode.
3D mode (current standard): Septa are retracted or absent. All cross-plane LORs are accepted,
dramatically increasing sensitivity (5–10× over 2D) at the cost of a higher scatter fraction (40–
60%). Modern corrections (model-based scatter simulation, TOF information) handle this
effectively. All clinical whole-body PET/CT studies are acquired in 3D mode.
The raw PET data are stored either as sinograms (the histogram mode—each LOR is binned by
radial distance and angle, forming a Radon transform representation) or as list-mode data (each
coincidence event is recorded with its detector-pair identity and exact timestamp). List-mode
enables retrospective gating (cardiac, respiratory), flexible rebinning, and optimal TOF
reconstruction but requires more storage.
Corrections Required Before Image Reconstruction
A raw PET sinogram does not produce a quantitatively accurate image. Six major corrections
must be applied:
• Attenuation correction (AC): 511 keV photons are attenuated as they pass through
tissue (μ ≈ 0.096 cm⁻¹ in water). For a 20 cm diameter patient, the probability that both
photons escape unattenuated is e⁻²°×¹² ≈ 0.15—meaning 85% of true coincidences are
lost. In PET/CT, the CT data (converted from Hounsfield units to 511 keV attenuation
coefficients by bilinear scaling) provide the attenuation map (μ-map). In PET/MR, MR-
based tissue segmentation or atlas-based methods are used. AC is the single most
important correction for quantitative accuracy.
• Scatter correction: Model-based scatter simulation (SSS—single scatter simulation)
uses an initial attenuation map and activity distribution to estimate the scatter component,
which is then subtracted from the prompts sinogram. This is the standard approach in all
clinical TOF-PET/CT systems.
• Random correction: Delayed coincidence channel method (see above) or analytical
estimation from singles rates.
• Dead time correction: At high count rates, the detector system is unable to process each
event individually. Dead time increases the apparent loss of counts at high activity.
Modern scanners characterise dead time empirically and apply a count-rate-dependent
correction factor.
• Decay correction: For accurate quantitation, counts must be corrected back to the time
of injection (or the beginning of each frame), accounting for radioactive decay of the
tracer during acquisition. For ¹⁸F with T½ = 109.8 min, a 60-minute dynamic study would
require significant frame-by-frame decay correction.
• Normalisation: Each detector element has slightly different efficiency due to crystal-to-
crystal variations in light yield, photodetector gain, and electronic thresholds.
Normalisation scans (using rotating rod sources or blank scans) characterise these
variations and produce efficiency correction factors applied to each LOR.
Spatial Resolution and Sensitivity
PET spatial resolution is governed by three physical factors: positron range (irreducible,
isotope-dependent; ¹⁸F ≈ 0.6 mm, Ga-68 ≈ 2.9 mm), non-collinearity (≈ 0.0022 × D, where D is
detector ring diameter; ≍ 0.9 mm at D = 40 cm), and crystal element size (inter-crystal scatter
and depth-of-interaction effects). In practice, clinical whole-body PET/CT achieves 4–6 mm
FWHM, while dedicated brain or small-animal scanners achieve 1–3 mm FWHM with smaller
crystal elements.
Sensitivity (the fraction of emitted photons detected as true coincidences) depends on detector
solid angle and crystal stopping power. Modern clinical scanners with 15–26 cm axial field-of-
view (AFOV) achieve approximately 5–12 kcps/MBq. Long AFOV scanners (uEXPLORER,
Biograph Vision Quadra, AFOV 106–194 cm) achieve 40–175 kcps/MBq, enabling total-body
PET at substantially reduced injected activity or scan time.
Comparison of PET Crystal Scintillators
Property BGO GSO LSO / LYSO LaBr₃
Density (g/cm³) 7.13 6.71 7.4 5.29
Effective Z 74 59 66 46
Decay time (ns) 300 60 40 16
Light yield 8–9 10–15 25–32 63
(photons/keV)
Energy resolution (%) 10–12 8–10 10–12 2.8–3.5
TOF-capable No Marginal Yes (210–250 ps) Yes (< 200 ps)
Intrinsic activity None None Yes (¹⁷⁶Lu β/γ) None
MR compatibility No (PMT) No (PMT) Yes (with SiPM) Yes (with SiPM)
Primary clinical use Legacy systems Some All modern PET/CT Research /
SPECT/PET PET/MR
Image Reconstruction
After correction, the sinogram data are reconstructed into a 3D activity distribution. The analytic
method FBP (filtered back-projection) is computationally fast but amplifies noise at low count
rates. All modern PET/CT scanners use iterative reconstruction, most commonly OSEM
(Ordered Subsets Expectation Maximisation), which models the statistical nature of photon
counting (Poisson statistics) and the physical characteristics of the scanner (system matrix) to
iteratively converge on the most likely activity distribution given the measured data. TOF
information is incorporated directly into the system matrix (TOF-OSEM), further improving
SNR and convergence speed. Point-spread function (PSF) modelling within OSEM corrects for
scanner-specific resolution losses and can partially recover spatial resolution at the cost of Gibbs
artefacts at edges of high-contrast structures.
Worked Numerical Example
Worked Example: Random Coincidence Rate and Attenuation Factor
In a whole-body ¹⁸F-FDG PET/CT study, detector pair AB has a singles rate of N₁ =
800,000 cps and detector pair CD has N₂ = 600,000 cps. The coincidence window is τ =
6 ns.
Given (a) Calculate the random coincidence rate between detectors A and C.
(b) For a 20 cm diameter uniform patient section (μ = 0.096 cm ⁻¹), calculate the
probability that BOTH 511 keV photons from a central annihilation escape unattenuated
(path length = 10 cm each).
(a) Random rate formula: R = 2τ × N₁ × N₂
R = 2 × 6 × 10⁻⁹ s × 8 × 10⁵ cps × 6 × 10⁵ cps
R = 2 × 6 × 10⁻⁹ × 4.8 × 10¹¹
R = 2 × 288 = 576 cps
Working
(b) Attenuation of each photon over 10 cm path:
Transmission for one photon = e⁻(μ×x) = e⁻(0.096 × 10) = e⁻¹·²²⁶ ≈ 0.301
For BOTH photons to escape: P = 0.301 × 0.301 = 0.0906
Attenuation factor = e⁻(μ×L) where L = total path = 20 cm
P = e⁻(0.096 × 20) = e⁻¹·⁹² = 0.1496
(a) Random coincidence rate R = 576 cps
(b) Only ~15% of true coincidences from the centre of a 20 cm patient escape without
Answer
attenuation, confirming why attenuation correction is the most critical quantitative
correction in PET. Using the combined path (20 cm) formula: e⁻¹·⁹² ≈ 0.15.
IMPORTANT FACTS
Annihilation photon energy: 511 keV (= 0.511 MeV = rest-mass energy of electron)
β⁺ decay threshold: Mass difference between parent and daughter must exceed 2 × 0.511 MeV =
1.022 MeV
¹⁸F properties: T½ = 109.8 min, E₋₊ⁿₑₜ = 633.5 keV, β⁺ fraction = 96.9%, positron range in tissue ≈
0.6 mm FWHM
Ga-68 properties: T½ = 67.7 min, E₋₊ⁿₑₜ = 1899 keV, positron range ≈ 2.9 mm FWHM (limits
resolution vs ¹⁸F)
Coincidence timing window: 6–12 ns (BGO systems use 12 ns; LYSO/SiPM systems use 6 ns or
narrower)
Scatter fraction, 3D mode: 30–40% whole-body; corrected by single scatter simulation (SSS)
Random correction: Delayed coincidence channel method; R = 2τN₁N₂
Attenuation coefficient (μ) for 511 keV in tissue: ≈ 0.096 cm⁻¹; in water ≈ 0.096 cm⁻¹; in bone ≈
0.17 cm⁻¹
TOF localisation: Δx = c × Δt / 2; CTR 210–250 ps → Δx ≈ 3–3.75 cm; SNR gain √(D/Δx)
LYSO intrinsic radiation: ¹⁷⁶Lu natural isotope (2.6% abundance) emits β + 307 keV γ; creates a
low-level background at ≈ 88, 202, 307 keV; does not significantly affect clinical imaging
PET spatial resolution (clinical, whole-body): 4–6 mm FWHM; limited by positron range + non-
collinearity + crystal size
Non-collinearity blur: ≈ 0.0022 × D (D = detector diameter); ≍ 0.9 mm at 40 cm diameter
Energy window for 511 keV: 350–650 keV (BGO); 425–650 keV (LYSO); LYSO’s narrower
window rejects more scatter
QUICK RECALL
PET physics sequence: β⁺ decay → positron range → annihilation → 2 × 511 keV photons at ~180°
→ coincidence detection → LOR → sinogram → correction → OSEM reconstruction
Energy of each annihilation photon: always 511 keV (= mₑc²)
Minimum Q-value for β⁺ decay: > 1.022 MeV
¹⁸F T½: 109.8 min; Ga-68: 67.7 min; Rb-82: 76 sec; Cu-64: 12.7 hr
Positron range (limits resolution): ¹⁸F 0.6 mm < Ga-68 2.9 mm < Rb-82 5.9 mm
Non-collinearity: ± 0.25°; causes ~0.9 mm blur at 40 cm ring diameter; irreducible
Electronic collimation: defines LOR without lead; sensitivity 20–50× higher than SPECT
Coincidence window: 6–12 ns (LYSO shorter); true, scatter, random events
Random rate formula: R = 2τN₁N₂; scales with count rate squared
Scatter correction method: Single scatter simulation (SSS); scatter fraction 30–40% in 3D
Attenuation correction: CT-based μ-map (Bilinear HU → μ at 511 keV); most critical quantitative
correction
LYSO decay time: 40 ns (enables TOF); BGO: 300 ns (no TOF)
TOF formula: Δx = cΔt/2; CTR 200 ps → Δx = 3 cm; SNR gain √(D/Δx)
OSEM: Ordered Subsets Expectation Maximisation; iterative; models Poisson statistics; standard
reconstruction
Clinical resolution: 4–6 mm FWHM whole-body; 1–3 mm brain/small animal
SiPM advantage over PMT: MR-compatible, faster timing (≤300 ps), lower bias voltage, compact;
enables PET/MR and TOF
MNEMONIC – 6 corrections: "A SNORD" – Attenuation, Scatter, Normalisation, raNdom, Dead
time, Radioactive decay