Why SUV Is Not a Constant

Physics · 2 min read

Uptake time, glucose, calibration, reconstruction and lesion size all move the number

SUV divides the activity concentration in a region by the injected activity per kilogram of body weight. It corrects for dose and patient size, and nothing else. Everything between the syringe and the reconstructed voxel can shift it, so two SUVs are only comparable when they were measured the same way.

Patient and protocol

• Uptake time. Tumour FDG uptake keeps rising over the usual imaging window, so a late scan reads higher. EANM recommends 60 min (acceptable 55–75 min) and, for repeat studies, the same interval to within 10 min. In one multicentre test–retest study, patients whose uptake times differed by more than 15 min had SUVmax variability above 21%.
• Glucose. Plasma glucose competes with FDG. In a small glucose-loading study, raising glucose from about 92 to 158 mg/dL cut the mean SUV of liver metastases from 9.4 to 4.3. Record glucose, and state whether any SUV is glucose-corrected.
• Extravasation. A paravenous injection traps tracer at the site and lowers every SUV. Image the injection site if in doubt, and report it.
• Bookkeeping. Ignoring residual syringe activity underestimates SUV by a median of about 2%. A 10 min clock error between dose calibrator and scanner gives about 6%. Weigh patients on a calibrated scale.

Scanner and reconstruction

• Calibration. The scanner must be cross-calibrated against the dose calibrator used for patient doses: quarterly, and after any upgrade. In the EARL accreditation programme, 5% of calibration submissions showed an SUV bias above 10%.
• Reconstruction. Iterations, filtering, point-spread-function modelling and time of flight all change SUVmax. In patients, the newer EARL2 standard gave SUVs 23–30% higher than EARL1.
• Partial volume. With 4–7 mm resolution, objects under about 3 cm read low. A 2 cm sphere recovers only 65–85% of its true value, depending on smoothing.
• Breathing. Motion can underestimate SUV of small lesions near the diaphragm by about 25% or more.

Harmonisation and real change

EARL accredits scanners against phantom recovery targets so that SUVs agree across sites. Use the EARL-compliant reconstruction for numbers, and a sharper one for reading if you wish. Keep serial scans on the same scanner, protocol and reconstruction.

With a careful protocol, tumour SUV has a within-subject coefficient of variation of about 10%. A fall of more than 25% or a rise of more than 33% is unlikely to be measurement noise.