Residency · Residency · Nuclear Medicine

PET Physics and Instrumentation

Positron Emission and Annihilation

Positron Decay Physics

Proton-rich radionuclides achieve stability through beta-plus decay, in which a proton converts into a neutron, a positron, and a neutrino. The emitted positron travels a short distance through surrounding tissue, gradually losing kinetic energy through collisions and near-collisions with atoms along its path. When the positron reaches thermal energies, it encounters an electron, and the two particles annihilate. This annihilation converts their combined rest mass into energy, producing two 511 keV photons emitted at approximately 180 degrees from one another. The simultaneous, back-to-back emission of these two photons is the physical principle that makes PET imaging possible.

Positron Range

The distance a positron travels from its point of emission to the site of annihilation is called the positron range, and it sets a fundamental limit on PET spatial resolution. Higher-energy positrons travel farther before annihilating, so the positron range varies among radionuclides. F-18 has the shortest mean range at roughly 0.6 mm (maximum about 2.4 mm), giving it the best intrinsic spatial resolution of the common PET isotopes. Ga-68 has a mean range of approximately 2.9 mm, producing slightly lower resolution. Rb-82, with its high positron energy, has a mean range of about 5.9 mm, resulting in significantly lower resolution. O-15 falls in between at roughly 3.0 mm mean range.

Non-Collinearity

The two annihilation photons are not emitted at exactly 180 degrees apart. Because the positron-electron pair retains some residual momentum at the moment of annihilation, the photons deviate by approximately 0.5 degrees from perfect back-to-back alignment. This non-collinearity introduces spatial blurring that increases with the diameter of the detector ring, contributing roughly 2 mm of resolution degradation in typical clinical PET scanners.

Coincidence Detection

Principles

PET exploits coincidence detection to localize annihilation events without physical collimation. When two opposing detectors simultaneously register 511 keV photons within a narrow coincidence timing window (typically 4 to 12 nanoseconds), the system records an event along the line of response (LOR) connecting those two detectors. The annihilation must have occurred somewhere along this line. Because no physical collimator is needed -- the geometry is defined electronically -- PET achieves dramatically higher sensitivity than SPECT, where the vast majority of photons are absorbed by the collimator.

Types of Coincidence Events

Not all detected coincidences represent valid spatial information. True coincidences occur when both photons from the same annihilation event reach their respective detectors without scattering, providing accurate positional data. Scattered coincidences arise when one or both photons undergo Compton scattering before detection, altering their trajectories and assigning the event to an incorrect LOR. Random (accidental) coincidences occur when two photons from entirely separate annihilation events happen to be detected within the coincidence timing window, again producing incorrect spatial information. The random coincidence rate is proportional to 2 x tau x R1 x R2, where tau is the timing window width and R1 and R2 are the singles count rates at the two detectors. Narrowing the timing window reduces the random rate. When more than two events are detected within a single timing window (multiple coincidences), the event is typically discarded because the correct pairing cannot be determined.

Prompt and Delayed Coincidence Windows

To correct for random coincidences, modern PET systems employ a delayed window technique. A second coincidence timing window, offset in time from the prompt window, measures the rate of chance coincidences. Since this delayed window cannot contain true coincidences, the counts it registers represent a direct estimate of the random rate. Subtracting the delayed window counts from the prompt window counts yields an estimate of true coincidences plus scattered events.

PET Detector Technology

Scintillator Materials

The choice of scintillator crystal has evolved significantly over the history of PET. BGO (bismuth germanate) was the original standard, valued for its high density (7.13 g/cm^3) and excellent stopping power for 511 keV photons. However, BGO has a slow scintillation decay time of approximately 300 ns, which limits timing performance. LSO (lutetium oxyorthosilicate, cerium-doped) revolutionized PET with its fast decay time of about 40 ns, high light output, and good stopping power, though it contains Lu-176 which produces a low-level background radioactivity. LYSO, with a very similar composition and performance to LSO, is equally widely used. The speed advantage of LSO and LYSO over BGO is critical because it enables time-of-flight PET and provides better energy resolution for improved scatter rejection. Modern PET crystals are typically cut to about 4x4 mm face dimensions and 20 to 25 mm depth.

Detector Block Design

The traditional PET detector consists of a block of scintillator crystals arranged in an array (such as 8x8 crystals) optically coupled to four photomultiplier tubes. Anger-type logic determines which crystal absorbed the photon based on the relative light distribution across the PMTs. In modern digital PET systems, this block architecture has been replaced by one-to-one coupling, where each individual crystal is paired with its own silicon photomultiplier.

Digital PET (SiPM-Based)

Silicon photomultipliers have transformed PET detector technology. With each crystal individually coupled to a SiPM, digital PET systems achieve superior timing resolution (below 250 picoseconds), better energy resolution, compact detector design, MRI compatibility (since SiPMs are unaffected by magnetic fields), and improved count rate performance. Digital PET with SiPM detectors is now the standard for most newly installed clinical PET/CT systems.

Time-of-Flight PET (TOF-PET)

Principle

Time-of-flight PET measures the tiny time difference between the arrival of the two annihilation photons at their respective detectors. Since light travels at a known speed, this time difference can be converted into a spatial offset, localizing the annihilation event to a specific segment of the LOR rather than treating the entire line as equally probable. The localization precision depends on the timing resolution: Delta_x = c x Delta_t / 2. A timing resolution of 500 picoseconds localizes the event to approximately 7.5 cm, while 200 picoseconds narrows this to about 3 cm. Importantly, TOF does not directly improve the spatial resolution of the final image but substantially improves the signal-to-noise ratio, since reconstruction need only distribute each event over a small segment of the LOR rather than its entire length.

Clinical Benefits

The signal-to-noise improvement from TOF-PET is proportional to the square root of (D/Delta_x), where D is the patient's diameter. This means the greatest benefit occurs in large patients, where the longer LORs and higher attenuation make conventional PET most challenging. TOF-PET enables lower-dose imaging or shorter scan times while maintaining image quality, accelerates convergence of iterative reconstruction, and improves lesion detection, particularly in obese patients.

PET Data Acquisition and Correction

2D vs. 3D Acquisition

Early PET scanners used inter-plane septa (thin lead or tungsten rings between detector rows) to restrict oblique lines of response, operating in 2D mode. This reduced scatter and random coincidences but also reduced sensitivity. Modern PET scanners operate exclusively in 3D mode, with no septa, accepting all oblique LORs and achieving roughly five times the sensitivity of 2D mode. The trade-off is that 3D acquisition requires more robust scatter and randoms corrections.

Attenuation Correction

Accurate attenuation correction is critical for quantitative PET because 511 keV photons undergo substantial attenuation as they pass through the body. At 511 keV, the half-value layer in soft tissue is approximately 7.2 cm. Unlike SPECT, where attenuation depends on the depth of the source, PET attenuation depends only on the total path length through the patient between the two detectors -- regardless of where along the LOR the annihilation occurred. CT-based attenuation correction (CTAC) is the standard method, using a low-dose CT scan to generate an attenuation map that is scaled from CT energies to 511 keV through bilinear conversion. Artifacts from CTAC include respiratory motion misregistration, metallic implant overcorrection, and intravenous contrast effects. Without attenuation correction, central structures appear artificially cold relative to the periphery.

Scatter Correction

In 3D PET, scattered events can represent 30 to 40% of all detected coincidences, making robust scatter correction essential. The most widely used algorithm is single scatter simulation (SSS), a model-based approach that estimates the scatter contribution from the combined emission and attenuation data.

Dead Time and Normalization

Dead time correction compensates for count losses that occur at high activity levels when the detector electronics cannot process every event. Detector normalization corrects for intrinsic variations in crystal efficiency and geometric factors across the detector ring. Both corrections are applied during data processing.

Decay Correction

Because the radionuclide's activity decreases during the scan, later bed positions would show reduced counts even if the activity distribution were uniform. Decay correction adjusts all data back to a reference time (typically the injection time) to ensure uniform quantification across the entire study.

PET Image Reconstruction

Iterative Reconstruction

OSEM is the standard reconstruction algorithm for PET, following the same iterative principles used in SPECT. The system model incorporated during reconstruction includes detector geometry, attenuation, scatter, random coincidences, and TOF information. Typical clinical parameters are 2 to 4 iterations with 8 to 21 subsets. Point spread function (PSF) modeling can be included for resolution recovery, analogous to collimator-detector response modeling in SPECT.

SUV Quantification

The Standardized Uptake Value (SUV) is the primary quantitative metric in clinical PET. It is calculated as the tissue radioactivity concentration (in kBq/mL) divided by the injected dose normalized to the patient's body weight. SUVmax represents the single hottest voxel within a region of interest and is the most commonly reported value. SUVmean is the average within a defined ROI. SUVpeak is the average within a 1-cm^3 sphere centered on the hottest voxel, and SULpeak is a lean-body-mass-corrected variant. While SUV is invaluable for clinical decision-making, it is influenced by numerous factors including patient weight, uptake time, blood glucose level, reconstruction parameters, and scanner calibration, limiting its reliability for inter-institutional comparisons.

PET vs. SPECT Comparison

<image>A detailed technical illustration of PET coincidence detection showing a ring of detector elements surrounding a patient. Illustrate a positron emitted from a tumor, traveling a short distance (positron range), then annihilating with an electron to produce two 511 keV photons at 180 degrees. Show the photons traveling to opposite detectors, with the line of response (LOR) drawn between them. Include insets showing the three types of coincidence events: true (both photons from same event reach correct detectors), scattered (one photon changes direction before detection, wrong LOR assigned), and random (two photons from different events detected within timing window).</image>

<image>An explanatory diagram of time-of-flight PET showing how measuring the time difference between photon arrivals at opposing detectors localizes the annihilation point along the LOR. Show two scenarios: conventional PET (event could be anywhere along the LOR, equal probability) versus TOF-PET (probability concentrated near the actual event location, shown as a Gaussian distribution along the LOR). Include timing resolution values (e.g., 500 ps vs. 200 ps) and corresponding spatial localization precision.</image>

<image>A side-by-side comparison of PET detector technology evolution: (1) traditional BGO block detector with 4 PMTs reading an 8x8 crystal array via Anger logic, (2) modern digital PET detector with LSO/LYSO crystals individually coupled to SiPMs in a 1:1 configuration. Label the components and include a performance comparison table showing timing resolution, energy resolution, and count rate capability improvements from analog to digital PET.</image>

Clinical Pearls

PET has fundamentally higher sensitivity than SPECT by a factor of roughly 100, because electronic collimation is vastly more efficient than physical collimation. This sensitivity advantage is why PET can detect smaller lesions and provide more reliable quantification.

Time-of-flight PET delivers its greatest benefit in large patients, where the SNR improvement is proportional to patient diameter. This is precisely the patient population where conventional PET image quality suffers most from attenuation and scatter.

F-18 provides the best intrinsic PET spatial resolution because of its short positron range of about 0.6 mm. Rb-82, with its much longer range of approximately 5.9 mm, produces inherently lower-resolution images, which is why cardiac Rb-82 PET images appear smoother than FDG studies.

Always inspect the CT attenuation correction maps for artifacts before finalizing PET interpretation. Respiratory misregistration, metallic implants, and intravenous contrast can all create false findings that may mimic or obscure true pathology.

SUV, despite its name, is not truly standardized. It varies with uptake time, blood glucose, patient weight, reconstruction method, and scanner calibration, which is why trending SUV values is most reliable when performed on the same scanner with consistent protocols.

Digital PET systems with SiPM detectors are now the standard for new installations, offering timing resolution below 250 picoseconds, improved energy resolution, and higher count rate performance compared to older PMT-based systems.

In 3D PET mode, which all modern scanners use exclusively, the scatter fraction reaches 30 to 40% of all coincidences. Robust scatter correction through single scatter simulation or similar algorithms is essential for quantitative accuracy.

References

• Cherry SR, Sorenson JA, Phelps ME. Physics in Nuclear Medicine. 4th ed. Elsevier; 2012. Chapter 18.
• Surti S, Karp JS. Advances in time-of-flight PET. Phys Med. 2016;32(1):67-72.
• Vandenberghe S, et al. Recent developments in time-of-flight PET. EJNMMI Phys. 2016;3:3.
• Boellaard R, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging. 2015;42(2):328-354.

Read this lecture as Markdown