What is PET reconstruction?
PET reconstruction is the mathematical process of converting millions of detected coincidence events into cross-sectional images representing the three-dimensional distribution of radiotracer activity.
Each coincidence event defines a line of response along which the annihilation occurred. Reconstruction algorithms use the complete set of these lines to estimate the activity distribution that most likely produced the measured data.
PET reconstruction converts coincidence event data into images using iterative algorithms that model physical processes to produce accurate three-dimensional tracer maps.
Modern PET systems use iterative reconstruction methods that model attenuation, scatter, detector response, and time-of-flight information to produce quantitatively accurate images.
Understanding the physics
When a PET scanner records a coincidence event, it does not know the exact annihilation point. It only ‘knows’ that it occurred somewhere along the line connecting two detectors. Over the course of an acquisition, millions of such lines of response are collected.
Reconstruction aims to determine the tracer distribution that best explains this dataset.
Early PET systems used filtered back projection, similar to SPECT. However, modern PET almost universally uses iterative reconstruction algorithms such as OSEM (Ordered Subsets Expectation Maximisation).
Iterative reconstruction begins with an initial estimate of tracer distribution. From this estimate, the system predicts what the projection data should look like. It then compares the predicted data with the measured coincidence data and updates the estimate to reduce the difference. This process is repeated multiple times.
Modern PET reconstruction incorporates modelling of:
Photon attenuation
Scatter correction
Detector geometry
Time-of-flight information
Incorporating time-of-flight narrows positional uncertainty along each line of response, improving signal-to-noise ratio and accelerating convergence.
The number of iterations and post-reconstruction smoothing influence image appearance. Increasing iterations improves contrast but if modelled too tightly can increase noise.
Where this matters clinically
Reconstruction parameters directly affect lesion detectability, image noise, and standardised uptake value (SUV) measurements. Understanding reconstruction principles helps explain differences between scanners and protocols.