Quantification of inter-observer variation before and after an educational intervention in MRgRT

Abstract

Purpose or Objective Accurate image registration is essential for magnetic resonance (MR) guided online treatment adaptive workflows (MRgRT). On-treatment MR-computed tomography (CT) image registration is a novel method for radiographers (RTTs) familiar with more established imaging platforms such as conebeamCT. In treatment sites demonstrating significant interfraction anatomical variation, such as the cervix, image registration can be particularly challenging. This work aims to validate the use of a guide developed by RTTs to improve interobserver variation for on-treatment image registration for cervix plan of the day MRgRT. Material and Methods Off-line, retrospective soft tissue image registration was undertaken by 4 RTTs for 10 cervical cancer patients undergoing radiotherapy on an ethics-approved imaging study. RTTs used the cervix to guide the registration of T2w MR images to the planning CT in Monaco (v5.19.05 Research, Elekta, Stockholm, Sweden). Resulting translations and RTT confidence scores for each registration were used to determine inter-observer variations. The exercise was repeated following the creation of a registration guide (including a case study atlas). This guide was developed collaboratively between RTTs and diagnostic MR radiographers, in consultation with clinicians, to identify the most relevant anatomic regions for registration. The guide was used as a reference for repeat registrations several weeks after the initial exercise. The mean, standard deviation (SD) and range of absolute difference among the three observers in right-left (RL), superior-inferior (SI) and anterior-posterior (AP) directions for each registration were calculated. Results Less variability was observed for MR-CT registrations when radiographers referenced a detailed registration protocol compared to matching the cervix alone. The interobserver mean, SD and range improved for all directions. Details can be found in table 1. Comparison by paired t-test of results before and after the use of the registration guide showed statistically significant differences for both the SI (p = 0.003) and AP directions (p = 0.001). Additionally, RTT confidence in registrations, using a Likert scale (1-5), improved from a mean score of 3.2 prior to the intervention to 3.9 following the use of the registration protocol. Conclusion This study has demonstrated the importance for radiographers operating in MGgRT to develop and evaluate robust consensus driven image registration procedures. These protocols aim to minimise uncertainty in image matching and ensure consistent and accurate treatment.