Evaluation and Optimization of a New PET Reconstruction Algorithm, Bayesian Penalized Likelihood Reconstruction, for Lung Cancer Assessment According to Lesion Size.

Abstract

<b>OBJECTIVE.</b> The purpose of this study was to characterize the Bayesian penalized likelihood (BPL) reconstruction algorithm in comparison with an ordered subset expectation maximization (OSEM) reconstruction algorithm and to determine its optimal penalization factor (expressed as a beta value) for clinical use. <b>MATERIALS AND METHODS.</b> FDG PET/CT scans of 46 patients with lung cancer were reconstructed using OSEM and BPL with beta values of 200, 300, 400, 500, and 1000. The liver signal-to-noise ratio, mean standardized uptake value (SUV_mean) of the liver, and maximum standardized uptake value (SUV_max) and SUV_mean of the cancers were measured. Tumors were categorized into three size groups, and the percentage difference in the tumor SUV_max between OSEM and BPL with a beta value of 200 as well as the percentage difference in the SUV_max between BPL with a beta value of 200 and BPL with a beta value of 1000 were calculated. Image quality was assessed by visual scoring. <b>RESULTS.</b> BPL showed a significantly higher liver signal-to-noise ratio than OSEM, except for BPL with a beta value of 200. The liver SUV_mean showed no statistical difference among all algorithms. The SUV_max and SUV_mean of tumors decreased as the beta value increased. BPL with a beta value of 200 produced a significantly higher tumor SUV_max than did OSEM (<i>p</i> < 0.01), and BPL with a beta value of 400, 500, or 1000 produced a significantly lower tumor SUV_max than did OSEM (<i>p</i> < 0.01). Visual analysis showed the highest and lowest scores for BPL with beta values of 500 and 200, respectively. In the small size group, the percentage difference in the SUV_max between OSEM and BPL with a beta value of 200 and the percentage difference in the SUV_max between BPL with a beta value of 200 and BPL with a beta value of 1000 were significantly larger than that in the other size groups (<i>p</i> < 0.01). <b>CONCLUSION.</b> The BPL algorithm improves image quality without compromising image quantification. A beta value of 500 appeared to be optimal in this study. Smaller tumors were more influenced by BPL.