Decoupling the dosimetric impact of setup errors and anatomical deformations in image-guided radiotherapy using diagnostic-quality FBCT and glmmLasso regularization

Abstract

Objective. Traditional radiotherapy dosimetric analyses are often limited by cone-beam CT artifacts and correlated longitudinal data mismanagement. This study uses artifact-free fan-beam CT (FBCT) and a generalized linear mixed Lasso (glmmLasso) model to decouple interfraction setup errors and anatomical deformations within a simulated image-guided radiotherapy workflow, providing quantitative support for future site-specific ART research. Approach. We retrospectively analyzed 84 patients (head and neck [H&N], thorax, pelvis; n = 28 each). Weekly diagnostic-quality FBCTs served as ground truth. To isolate error sources, we recalculated 1 260 dose distributions under three scenarios: isolated setup errors, isolated deformations, and combined effects. A glmmLasso model, using patient-specific random intercepts and L1 penalization, identified the primary geometric drivers of dosimetric variation. Main Results. H&N exhibited minimal setup deviations (< 2.0 mm); dosimetric variance was driven by inter-patient baseline heterogeneity (ICC > 0.73) rather than interfraction deformations. Thoracic setup uncertainties (mean≈4.0 mm) reduced minimum target dose up to 50% for displacements ⩾ 5 mm, but overall variance was primarily governed by non-rigid deformations (marginal R2 = 0.423). Pelvic anatomy demonstrated pronounced variability due to bladder filling (DSC: 0.2–0.9), and the coupled scenario further amplified relative deviations in bladder and femoral-head dose metrics. Significance. Utilizing FBCT and high-dimensional feature selection, we decoupled site-specific drivers of dosimetric variation. Quantifying these coupled effects establishes a robust, validated foundation for tailored potential factors to inform ART decision-making.