Interobserver image registration variability impacts on stereotactic arrhythmia radioablation (STAR) target margins

Background and purpose. To determine the interobserver variability in registrations of cardiac computed tomography (CT) images and to assess the margins needed to account for the observed variability in the context of stereotactic arrhythmia radioablation (STAR).Materials and methods.STAR targets were delineated on cardiac CTs for fifteen consecutive patients. Ten expert observers were asked to rigidly register the cardiac CT images to corresponding planning CT images. Registrations all started with a fully automated registration step, followed by manual adjustments. The targets were transferred from cardiac to planning CT using each of the registrations along with one consensus registration for each patient. The margin needed for the consensus target to encompass each of the observer and fully automated targets was measured.Results.A total of 150 registrations were evaluated for this study. Manual registrations required an average (standard deviation) of 5 min, 55 s (2 min, 10 s) to perform. The automated registration, without manual intervention, required an expansion of 6 mm to achieve 95% overlap for 97% of patients. For the manual registrations, an expansion of 4 mm achieved 95% overlap for 97% of the patients and observers. The remaining 3% required expansions from 4 to 9 mm. An expansion of 3 mm achieved 95% overlap in 88% of the cases. Some patients required larger expansions compared to others and small target volume was common among these more difficult cases. Neither breath-hold nor target position were observed to impact variability among observers. Some of the observers required larger expansions compared to others and those requiring the largest margins were not the same from patient to patient.Conclusion.Registration of cardiac CT to the planning CT contributed approximately 3 mm of uncertainty to the STAR targeting process. Accordingly, workflows in which target delineation is performed on cardiac CT should explicitly account for this uncertainty in the overall target margin assessment.