Purpose or Case Report: We aim to evaluate qualitative and quantitative image quality of low-dose PET-MRI in young patients.
Methods & Materials: 25 PET/MRI scans from 14 patients under the age of 5 years were analyzed. Mean age was 2.78 years. Imaging protocol included 4 minute PET acquisition per bed position with a mean of 4.1 MBq/kg activity of FDG. Simulated half-dose images were created by removing events from the PET file list. In addition to ordered subset expectation maximization (OSEM) reconstructions, PET data was reconstructed with Bayesian penalized likelihood reconstruction (Q.Clear, β = 350) for full and half-dose images.

MIP images from 4 reconstructions were viewed and rated by 2 radiologists based on image quality, noise, and sharpness on a 5 point Likert scale and artifacts on a 3-point Likert scale. Likert scale data was analyzed with Kruskall-Wallis test. Post-hoc pairwise analysis with Dunn test was performed with statistically significant differences (p<0.05). Weighted Cohen's kappa was calculated for interobserver agreement.

SUV analysis included the SUV_max, SUV_avg, signal-to-noise (SNR) for liver, and contrast-to-noise (CNR) for lesions. SUVs were analyzed by Friedman’s test and post-hoc analysis with Nemenyi test.
Results: Likert data showed no statistical difference for image quality (chi² = 5.1394, p= 0.16), sharpness (chi² = 2.81, p = 0.24) Kruskal-Wallis test showed a significant difference for noise (chi² = 20.187, df = 3, p = 0.0001). Post-hoc Dunn test showed half-dose scans were subjectively noisier than the full-dose scans. The Q.clear did not subjectively decrease noise when compared to the OSEM reconstructions. No artifacts were detected. Moderate agreement was shown (Κ= 0.411).

Liver SUVmax was higher in the half-dose scans (p from 4.9 x10^-14 to 1.9 x10^-5). However, liver SUV_avg showed a statistically significant difference in only the half-dose OSEM vs. full-dose Q.Clear with bias of only 10%(p = 1.6 x10^-5). Liver SNR decreased with half-dose and Q.Clear (p-value from 3.8 x10^-14 to 0.024). Lesion SUVavg and SUVmax were significantly increased in the Q.Clear than the OSEM (p-value ranging 1.5 x 10^-7 to 0.007) in both dose reconstructions. CNR was decreased in the half-dose (p from 6.1 x 10^-5 to 0.006) and did not improve with Q.Clear (p from 0.65 to 0.91).


Conclusions: Simulated half-dose show qualitatively diagnostic but nosier images. Quantitatively, Q.Clear did not improve SNR or CNR. Future directions could explore optimization of Q.Clear’s β value.