A prospective study using a spirometry-based system on the positional reproducibility of anatomical landmarks and tumour-tracking accuracy.

Abstract

This prospective study evaluated the positional reproducibility of anatomical landmarks and estimated planning target volume (PTV) margins using a spirometry-based system during deep inspiration breath-hold (DIBH), and evaluated the system's potential as a surrogate for dynamic tumour tracking (DTT). The study comprised two components, each involving 10 patients and utilising a spirometry-based system. Part A evaluated inter- and intra-fractional variations at 12 bronchial bifurcation landmarks and estimated PTV margins based on vertebral- and carina-based registrations. Part B assessed six 4D tumour position prediction models using varying ratios of spirometry- and surface-based inputs. The root-mean-square error (RMSE) was used to evaluate the prediction accuracy over two time intervals. For short-term evaluation, 20 s of data were used for model training, and the subsequent 50 s for validation. For long-term evaluation, a separate 70-second dataset was used. In Part A, mean inter- and intra-fractional variations across the 12 landmarks were 4.2 ± 2.0 mm and 2.9 ± 2.0 mm, respectively. PTV margins remained < 5 mm for both registrations, except in the superior-inferior direction of the left anteromedial segment. In Part B, no significant RMSE differences were observed in short-term predictions. For long-term predictions, the spirometry-only and combined-input models showed significantly lower RMSE than the surface-only model (P = 0.049 and P = 0.021, respectively). Spirometry-based DIBH demonstrated acceptable positional reproducibility; however, individualised PTV margins may be necessary for specific regions. Spirometry-based prediction remained robust during free breathing, supporting its utility as a surrogate for DTT.