PET & SPECT Performance & Reconstruction
Resolution limits, coincidence physics, NECR, and iterative reconstruction, the working layer
Quick Answer
SPECT resolution is limited chiefly by the collimator (resolution degrades with source-to-collimator distance) and combines with intrinsic detector resolution in quadrature. PET needs no physical collimator (coincidence detection) but is bounded by positron range, photon non-collinearity, and detector element size; its signal is the true coincidences against scatter and randoms, summarized by the noise-equivalent count rate (NECR). Iterative reconstruction (OSEM) with PSF and time-of-flight (TOF) modeling, plus attenuation/scatter/partial-volume corrections, turn raw data into quantitative images.
SPECT — Resolution & Sensitivity
System resolution combines detector (intrinsic) and collimator (geometric) terms in quadrature:
R_system = √(R_intrinsic² + R_collimator²)
- The collimator dominates and degrades with distance from the collimator face — so keep the detector close to the patient.
- Resolution–sensitivity trade-off: high-resolution collimators accept fewer photons (lower sensitivity); high-sensitivity collimators blur. A converging (fan/cone) collimator or CZT/dedicated cardiac geometry can improve one without fully sacrificing the other.
- Pinhole collimation magnifies for high-resolution small-organ imaging (thyroid, parathyroid, small-animal).
PET — Intrinsic Resolution Limits
Even with electronic collimation, PET resolution is bounded by physics:
| Factor | Effect |
|---|---|
| Positron range | Positron travels before annihilation — larger for high-energy emitters (Rb-82, Ga-68) than F-18 |
| Photon non-collinearity | The two 511-keV photons are ~0.5° off 180°, blurring by an amount that grows with detector-ring diameter (~½ FWHM contribution at ~80 cm) |
| Detector element size | Contributes ~(element size)/2 to intrinsic resolution |
| Depth-of-interaction | Parallax blurring off-center in the field |
These set a practical whole-body PET resolution around a few millimeters — the floor beneath which no reconstruction can recover true detail.
PET — Coincidences: True, Scatter, Random
- True coincidence: both photons from one annihilation, unscattered — the signal.
- Scatter coincidence: one/both photons Compton-scattered — mispositioned; removed by scatter correction.
- Random (accidental) coincidence: two photons from different annihilations within the timing window — increase with the square of activity; corrected by the delayed-window technique.
- Noise-equivalent count rate (NECR) = T²/(T + S + R) — the effective signal-to-noise metric that peaks then falls as activity rises (randoms/dead-time dominate at high activity), guiding optimal injected activity.
Time-of-Flight (TOF)
Fast detectors measure the arrival-time difference of the two photons and localize the annihilation along the line of response, narrowing the probable position. TOF improves signal-to-noise (effective sensitivity gain that scales with patient size) — most valuable in larger patients.
Image Reconstruction
- Filtered back-projection (FBP): fast, linear, but streak-artifact-prone and noisy — largely historical for clinical use.
- Iterative (OSEM): models the acquisition (system matrix) and iterates toward the image; subsets × iterations trade convergence against noise. Incorporates:
- PSF / resolution recovery — models spatially varying blur (sharper images; can introduce edge "Gibbs" overshoot and change SUV).
- TOF (PET) — improves SNR/convergence.
- Reconstruction settings change SUV, so quantitative comparison requires matched reconstruction (or an EARL-harmonized reconstruction).
Quantitative Corrections
| Correction | Purpose | Failure mode |
|---|---|---|
| Attenuation (CT map) | Recover photons absorbed en route | CT–emission misregistration → false cardiac defects |
| Scatter | Remove mispositioned scattered photons | Loss of contrast if uncorrected |
| Randoms (PET) | Subtract accidental coincidences | Elevated background at high activity |
| Dead time | Correct count-rate losses | Underestimation at high rates |
| Partial-volume | Recover activity in small structures | Small-lesion SUV underestimated (recovery coefficient < 1 below ~2–3× resolution) |
The Partial-Volume Effect & Recovery Coefficient
Structures smaller than ~2–3× the system resolution (FWHM) have their activity underestimated (spill-out) and background overestimated (spill-in). The recovery coefficient (measured/true activity) falls below 1 as size decreases — the physical reason a small avid nodule can show a falsely low SUV, and a caution in serial small-lesion quantitation.
NEMA Performance Standards
NEMA NU-2 (PET) and NU-1 (SPECT) standardize measurement of spatial resolution, sensitivity, scatter fraction, NECR, count-rate performance, and image quality, enabling scanner comparison and acceptance testing.
Board-Level Synthesis
SPECT resolution is collimator-limited and degrades with distance — combine intrinsic and collimator terms in quadrature and keep the detector close. PET floor is set by positron range, non-collinearity (~0.5°, worse at larger ring diameter), and detector element size — no reconstruction recovers detail below it.
PET signal is true coincidences against scatter and randoms (randoms ∝ activity²); NECR = T²/(T+S+R) peaks then falls with activity, guiding optimal dose. TOF localizes the annihilation along the line of response, improving SNR most in larger patients.
Iterative reconstruction (OSEM) with PSF/resolution recovery and TOF — plus attenuation/scatter/randoms/partial-volume corrections — makes images quantitative, but changes SUV, so serial comparison needs matched (or EARL-harmonized) reconstruction. The partial-volume effect underestimates SUV in structures below ~2–3× FWHM (recovery coefficient < 1) — the physics behind falsely low small-lesion SUV and the CT–emission misregistration artifact behind spurious cardiac defects.
Related Pages
- Physics: Radiation detection & instrumentation, SPECT/PET & hybrid imaging, physics of nuclear medicine (positron range).
- Reference: SUV harmonization (EARL); pitfalls: Pearls, pitfalls & normal variants.
Figure / Diagram Suggestions
- A PET resolution-budget diagram (positron range + non-collinearity + detector size).
- A NECR curve (true/scatter/random vs activity).
- A recovery-coefficient curve (SUV recovery vs lesion size / FWHM).
Self-Check
Q1. Why does SPECT resolution worsen as the patient is farther from the collimator?
Answer: The collimator geometric resolution degrades with source-to-collimator distance and dominates system resolution — so the detector should be kept as close to the patient as possible.
Q2. Name the three physical factors limiting PET spatial resolution.
Answer: Positron range, photon non-collinearity (~0.5°, worse with larger ring diameter), and detector element size (plus depth-of-interaction/parallax).
Q3. What is NECR and why does it matter for injected activity?
Answer: Noise-equivalent count rate = T²/(T+S+R); it peaks then falls as activity rises (randoms/dead-time), identifying the activity that optimizes effective signal-to-noise.
Q4. Below what size are lesion SUVs underestimated, and what quantity describes this?
Answer: Below roughly 2–3× the system resolution (FWHM), due to the partial-volume effect; the recovery coefficient (measured/true) falls below 1.
Evidence & sources
https://nuc.med/pet-spect-performance-reconstruction/ Report an issue