Detectability of small objects in PET/computed tomography phantom images with Bayesian penalised likelihood reconstruction.

Objectives: This study investigated the subcentimetre lesion detection gains of a Bayesian penalised likelihood reconstruction (BPLR) (Q.Clear, GE Healthcare, Milwaukee, USA) in PET/computed tomography (CT) phantom images and compares observer performance with ordered subset expectation maximisation (OSEM) reconstruction images (VUE Point HD, GE Healthcare).

Methods: Images were presented to three blinded experienced observers to identify lesions and assign confidence ratings. Responses were analysed using jackknife alternative free receiver operator characteristic (JAFROC) software. Phantom lesions (active and nonactive) were constructed using putty. Seventy nonactive and 93 (F) active lesions, with diameters of 3, 5 or 7 mm were suspended in active backgrounds at varying contrast ratios (2:1-32:1) within an National Electrical Manufacturers Association 2012 phantom. PET/CT images were acquired with a GE Discovery 710 and reconstructed using both BPLR (penalisation coefficient 400) and high-definition attenuation corrected (HDAC) OSEM (2 iterations, 24 subsets).

Results: Small but significant (P = 0.009) visual detection gains were seen for active lesions with BPLR [weighted JAFROC figure of merit (wJAFROC FOM) = 0.77] over OSEM (FOM = 0.74). When split by subset, these improvements were significant for 5 mm and lesion to background ratio of 8:1. No significant (P = 0.514) differences were seen for the identification of nonactive lesions of any size (BPLR FOM = 0.74 and OSEM FOM = 0.73).

Conclusions: Significant detection gains were demonstrated for small active lesions with BPLR over OSEM. Coupled with the significant increase in contrast-to-noise ratio, these results support the use of BPLR in the imaging of small active (≤7 mm) lesions but show no improvement with BPLR in the identification of true negative lesions.