'Bad-breath' rejection: quality control metrics for respiratory gating in PET/CT.

Objectives: To develop a metric for detecting anomalous breaths to be rejected from respiratory-gated PET acquisitions.

Methods: Histograms of the respiratory rates of patients exhibit skewed distributions that are not accounted for by symmetric cycle rejection techniques (e.g. ± 20% of mean respiratory rate). Overestimation of ‘bad-breaths’ results in fewer acquired counts in the reconstruction and increased noise in the image. Thus, we developed an outlier estimation technique that accounts for the skewness of the real respiratory cycle distributions to deduce meaningful limits for ‘bad-breath’ rejection. A cohort of 50 Rb-82 myocardial perfusion PET/CT studies at rest or stress was processed with the data-driven PeTrack algorithm [1] to derive respiratory gating triggers. Trigger event time intervals were examined to derive meaningful limits for respiratory-cycle rejection. For each acquisition respiratory cycles were normalized to the median and a log-transformation was performed to reduce the extent of skewness. An adjusted boxplot [2] of the distribution was produced for outlier estimation of the skewed distribution. Respiratory cycles that fell outside of the boxplot whiskers were rejected. In this approach the standard boxplot whiskers that are 1.5 times the interquartile range are modified using exponential factors that depend on the medcouple statistic of skewness [3]. This proposed method is referred to as ‘BBR’ hereafter. A symmetric method (‘SYM’) was also investigated with which respiratory cycles that are outside the limits of ± 20% of the mean respiratory rate were rejected. The rate of rejected respiratory cycles and the total acquisition time rejected for each study were recorded for BBR and SYM methods. Respiratory-gated image reconstructions were performed on one example acquisition to indicate the effect of each cycle rejection method on the noise characteristics as measured by SNR and CNR. The volume of the left-ventricle (LV) was also compared to evaluate residual motion blurring between the two approaches.

Results: The SYM approach resulted in a mean rejection rate (± SD) of 50 ± 19% of respiratory cycles, with per patient minimum and maximum rejection rates of 10% and 85%, respectively. In contrast, the BBR method exhibited a significantly lower mean rejection rate of 10 ± 7 % (p<0.0001 compared to SYM, two-sampled t-test) and minimum and maximum rejection rates of 1% and 30%, respectively. The mean rejected acquisition time observed using each method was 242 ± 69 s and 57 ± 36 s (scan duration was 480 s) for SYM and the BBR approaches, respectively. LV volume, SNR and CNR were computed for the end-expiration image of one example case. Mean increases of 11%, 66% and 92% were observed for LV volume, SNR and CNR for BBR approach compared to SYM. For the example case the SYM approach rejected 28% (177 s of data) of the respiratory triggers while only 0.9% (5 s of data) were rejected using BBR. Conclusions: An alternative approach for respiratory cycle rejection has been considered for phase-based respiratory gating in PET/CT. Results suggest that symmetric limits about the mean respiratory rate do not accurately account for the skewed nature of human respiratory intervals. The BBR method significantly reduced respiratory cycle rejection rates thus preserving a greater amount of the acquired data in the gated reconstructions leading to improved SNR and CNR.