Attenuation correction is a processing step in medical imaging that compensates for the body’s natural tendency to absorb and scatter radiation before it reaches the scanner’s detectors. Without it, images from PET and SPECT scans would show tissues deep inside the body as falsely dim, while structures near the surface would appear artificially bright. The correction works by mapping how dense each part of the body is, then mathematically adjusting the scan so that every region reflects its true level of activity.
Why Raw Scans Are Inaccurate
PET and SPECT imaging rely on detecting gamma ray photons emitted from inside your body. As these photons travel outward through tissue, bone, and air pockets, some are absorbed entirely and others bounce off course in a process called Compton scattering. A scattered photon changes direction and loses energy, which means it either misses the detector completely (registering as a lost count) or gets picked up in the wrong location, creating a false background signal.
The deeper an organ sits inside the body, the more tissue its photons have to pass through and the more signal gets lost. This creates a systematic bias: a lesion near the center of the chest, for example, would look less active than an identical lesion just under the skin. For scans used to measure metabolic activity or blood flow, that kind of error can mask disease or make healthy tissue look abnormal.
How CT-Based Correction Works
The most common method pairs the PET or SPECT camera with a CT scanner built into the same machine. The CT scan produces a detailed density map of the body, measured in Hounsfield units. Air registers at negative 1,000, water at zero, and dense cortical bone above positive 1,000. Software converts these density values into a number called a linear attenuation coefficient for each tiny volume of tissue, essentially quantifying how much that spot would block or scatter a photon at the energy levels used in the nuclear medicine scan.
The conversion uses a two-part linear model. For tissues less dense than water (like lung and fat), the software draws a straight line between the known values for air and water. For denser tissues (like muscle and bone), it draws a second line between water and cortical bone. Each pixel in the CT image gets mapped to its corresponding attenuation value, and the resulting correction map is applied to the PET or SPECT data during image reconstruction. The corrected image shows tracer uptake that reflects actual biology rather than how deep or surrounded by dense tissue a structure happens to be.
The Impact on Measurement Accuracy
Quantitative PET imaging depends on a metric called the standardized uptake value (SUV), which reflects how much tracer a given spot has absorbed relative to the rest of the body. Oncologists use SUV to gauge tumor aggressiveness and track treatment response, so even small errors matter.
Research comparing CT-based correction against older transmission-source methods found that SUV measurements in bone can be underestimated by roughly 2.4% to 5.9% depending on the CT settings used. At lower-energy CT settings (120 kVp), the maximum SUV error can reach about 6.8%. These percentages may sound modest, but in serial scans tracking whether a tumor is shrinking, a consistent 5% to 7% bias can obscure meaningful changes. The American College of Radiology requires accredited PET facilities to verify the accuracy of both their CT numbers and their SUV measurements as part of annual quality control testing.
MRI-Based Correction in PET/MR Systems
PET/MR scanners don’t have a built-in CT, which creates a unique challenge: standard MRI sequences are excellent at imaging soft tissue but produce almost no signal from bone. Since bone and air both appear dark on conventional MRI, the software can’t easily tell a dense skull from an air-filled sinus, and those two structures attenuate photons very differently.
To solve this, specialized pulse sequences called ultrashort echo time (UTE) and zero echo time (ZTE) capture signal from bone before it decays. On a ZTE image, soft tissue appears bright, cortical bone appears moderate, and air appears dark, giving the software enough contrast to distinguish all three. A hybrid approach combines ZTE data (for bone density estimation) with a separate Dixon MRI sequence (which separates water, fat, and air), producing a complete attenuation map without any radiation from a CT scan.
Artifacts From Metal Implants
Metal hardware like hip replacements, spinal rods, or dental fillings can distort the attenuation correction process. In CT-based systems, metal causes bright streaks on the CT image that inflate the local density values, potentially making nearby tissue appear more metabolically active than it really is. In MRI-based systems, the problem flips: metal creates a signal void that the software interprets as air, assigning a near-zero attenuation value to the surrounding region. Because the correction then assumes those tissues barely exist, the reconstructed PET image overestimates tracer activity in and around the implant.
The signal void on MRI is typically much larger than the implant itself, amplifying the error. This can also produce unexpected artifacts in areas far from the metal, with overestimation spreading horizontally outside the immediate artifact zone. Radiologists reading these scans factor in known implant locations to avoid misinterpreting artificially hot or cold spots as disease.
Radiation Dose From the CT Component
Because the CT scan used for attenuation correction adds ionizing radiation on top of the tracer injection, minimizing its dose is a practical concern. A standard low-dose CT for correction purposes typically runs at 50 to 100 milliamps, while ultra-low-dose protocols push below about 5 milliamp-seconds, delivering roughly 0.5 millisieverts of whole-body effective dose. For context, a full diagnostic chest CT delivers around 7 millisieverts. The attenuation correction CT doesn’t need to be diagnostic quality; it just needs to accurately distinguish tissue densities, so facilities can use substantially lower radiation settings.
Deep Learning Approaches
A newer strategy skips the CT or MRI entirely by using artificial intelligence to estimate tissue density directly from the uncorrected PET image. A deep convolutional neural network is trained on thousands of paired PET/CT datasets, learning the relationship between the pattern of tracer distribution in an uncorrected scan and the corresponding CT density map. Once trained, the network takes a single uncorrected PET image as input and generates a synthetic CT image, pixel by pixel, that can be used for attenuation correction.
This approach has been validated with the common cancer-imaging tracer FDG and produces quantitatively accurate corrected images. Its main appeal is enabling PET-only imaging without the added radiation of a CT scan or the cost and complexity of an MRI. It also opens possibilities for older standalone PET cameras that were never paired with a CT scanner, potentially extending the useful life of existing equipment.