Comparison of Q.Clear (regularized) reconstruction versus OS-EM for characterization of small lesions in oncologic PET/CT

Objectives Accurate characterization of small lesions in oncologic PET/CT may better inform clinical decision making and allow for earlier and more effective therapy. This goal is a challenging one in PET due to limited spatial resolution, noise, respiratory motion, and other technical factors. The Q.Clear reconstruction algorithm installed on the recently introduced GE Discovery IQ PET/CT ("IQ") (GE Healthcare, Milwaukee WI), is a regularized method that employs noise penalties to allow the resolution benefits of high numbers of iterations without the corresponding high noise. The purpose of this study is to determine whether Q.Clear improves the visual conspicuity and increases the SUVmax of lesions clinically deemed "too small to characterize" relative to traditional ordered subset expectation maximization (OS-EM) reconstruction.

Methods In this IRB-approved, HIPAA-compliant retrospective study, 30 patients were selected who underwent F-18 FDG PET/CT on the IQ between 4/15/2015 and 11/12/2015 and whose dictated clinical PET/CT reports described lesions as being "too small to characterize on PET", "below PET resolution", and the like. There were 15 female and 15 males with mean age 65 (range 29-83) years, with cancers of lung, bladder, breast, cervix, endometrial, esophageal, head and neck, lymphoma, and melanoma primaries. After i.v. injection of 0.4- 0.8 mC/kg FDG and a 60 min uptake period, patients were scanned for 2-4 min/ bed position per our standard protocol. Images were reconstructed with corrections for scatter, dead time, randoms, and attenuation, into a 128x128 matrix in a 50 cm FOV. OS-EM was performed with 2 iterations and 12 subsets. Q.Clear images were reconstructed per manufacturer default settings with a β parameter of 350. Q.Clear and OS-EM images were reviewed alongside the CT, and assessed for subjective target lesion conspicuity (0= not seen, 1= somewhat conspicuous, 2=very conspicuous), lesion SUVmax, and background SUVmean (mediastinal blood pool and liver). Differences were assessed using Student's T test.

Results Thirty-six lesions on 30 PET/CT's were analyzed: 24 lung nodules, 4 liver lesions, 2 bone lesions, 5 lymph nodes, and 1 adrenal nodule. Mean CT diameter was 7 mm (range 2-25 mm). Mean lesion SUVmax was 1.4 for Q.Clear and 1.4 for OS-EM. SUVmean of the liver and blood pool were 2.1 and 1.7, respectively, identical between reconstruction methods. On the OS-EM images, 20 lesions had a conspicuity score of 0, 13 lesions a score of 1, and 3 lesions a score of 2. For only one lesion did Q.Clear increase the conspicuity score above that seen on OS-EM; otherwise scores were unchanged (n=32) or decreased (n=3) relative to OS-EM. Q.Clear reconstruction overall resulted in a nonsignificant decrease in subjective conspicuity compared to OS-EM (mean score was 0.47 for Q.Clear and 0.53 for OS-EM, (p=0.32)).

Conclusions Q.Clear reconstruction neither significantly improved lesion conspicuity nor increased the measured SUVmax in lesions clinically interpreted as "too small to characterize" on PET/CT, relative to OS-EM. Further analysis will be performed varying Q.Clear β and other reconstruction parameters, and with larger lesion size. Additionally, fractional count images will be reconstructed to determine whether Q.Clear may be of value in increasing lesion conspicuity in situations of low counts.