Comparison of Bayesian penalized likelihood reconstruction versus OS-EM for characterization of small pulmonary nodules in oncologic PET/CT.

OBJECTIVE: To determine whether the recently introduced Bayesian penalized likelihood PET reconstruction (Q.Clear) increases the visual conspicuity and SUVmax of small pulmonary nodules near the PET resolution limit, relative to ordered subset expectation maximization (OS-EM). METHODS: In this institutional review board-approved and HIPAA-compliant study, 29 FDG PET/CT scans performed on a five-ring GE Discovery IQ were retrospectively selected for pulmonary nodules described in the radiologist's report as "too small to characterize", or small lung nodules in patients at high risk for lung cancer. Thirty-two pulmonary nodules were assessed, with mean CT diameter of 8 mm (range 2-18). PET images were reconstructed with OS-EM and Q.Clear with noise penalty strength β values of 150, 250, and 350. Lesion visual conspicuity was scored by three readers on a 3-point scale, and lesion SUVmax and background liver and blood pool SUVmean and SUVstdev were recorded. Comparison was made by linear mixed model with modified Bonferroni post hoc testing; significance cutoff was p < 0.05. RESULTS: Q.Clear improved lesion visual conspicuity compared to OS-EM at β = 150 (p < 0.01), but not 250 or 350. Lesion SUVmax was increased compared to OS-EM at β = 150 and 250 (p < 0.01), but not 350. CONCLUSION: In a cohort of small pulmonary nodules with size near an 8 mm PET full-width half maximum, Q.Clear significantly increased lesion visual conspicuity and SUVmax compared to our standard non- time-of-flight OS-EM reconstruction, but only with low noise penalization. Q.Clear with β = 150 may be advantageous when evaluation of small pulmonary nodules is of primary concern.

Duke Scholars

Author Brandon Augustus Howard Radiology, Nuclear Medicine

Author Timothy Garvey Turkington Radiology

Author Olga Gennadyevna James Radiology, Nuclear Medicine

Author Salvador Borges-Neto Radiology, Nuclear Medicine