Facilitating the mid-ventilation SABR lung technique with deformable image registration: retrospective study into contour accuracy.

Abstract

PurposeThe primary aim was to evaluate the accuracy of the generated Mid-Ventilation (MidV) Gross Tumour Volume (GTV) and Planning Target Volume (PTV) using three Deformable Image Registration (DIR) algorithms (Eclipse DIR, MIM DIR, Velocity DIR) and two starting phases for contour propagation (0% vs. 20%), compared with inter-observer variation (IOV).Methods27 stereotactic ablative radiotherapy (SABR) lung datasets were analysed. For each patient, three contours and a consensus reference contour was created. This reference contour was used to generate six DIR-based 4D contours (three algorithms with two starting phases). Performance was assessed using contour agreement metrics (mean distance to agreement [MDA)], Dice similarity coefficient [DSC], Hausdorff distance [HD], and volume ratio). Non-inferiority tests (one-sided, Dunnett-adjusted) evaluated DIR performance relative to IOV.ResultsDIR with a 20% starting phase outperformed 0% predominantly. Eclipse and Velocity DIR-20% met all published tolerances (MDA within 2 mm, DSC more than 0.8, and volume ratio within 10%) and were non-inferior to IOV. Secondary analyses showed improved consistency in MidV phase determination, MidV to Mid-position (MidP) centroid deviation (within 1 mm), 4D motion estimation (ConclusionDIR algorithms using a 20% starting phase demonstrated superior performance compared with a 0% starting phase. Commercial DIR algorithms can support MidV contouring for lung SABR when initiated from the 20% starting phase. DIR-20% produced contour accuracy comparable to IOV with robust performance across MidV subprocesses.