Epid-Based Setup Error Correction in Head and Neck Radiotherapy Publisher



Pursamimi M ; Ghorbani M ; Khosravi M ; Hosseini SM ; Shakeri A ; Mdletshe S
Source: Iranian Journal of Medical Physics Published:2026

Abstract

Introduction: This study introduces and evaluates a novel optimized averaging method for Electronic Portal Imaging Device (EPID)-based setup corrections, comparing it with conventional approaches. Material and Methods: A total of 28 head and neck cancer patients undergoing conformal radiotherapy were enrolled prospectively. Three EPID-based setup correction methods were compared: no averaging (Method 1), standard three-fraction averaging (Method 2), and a new optimized averaging method (Method 3). Setup errors were quantified in three dimensions, and Clinical Target Volume to Planning Target Volume (CTV-PTV) margins were calculated. Primary outcomes included systematic and random errors, CTV-PTV margin reduction, and the proportion of patients benefiting from each method. Results: Both averaging methods significantly reduced systematic and random errors compared to no averaging (p<0.05). Mean setup errors decreased to <1 mm in all directions for Methods 2 and 3. CTV-PTV margins were reduced from 5.7-7.9 mm to 3.6-3.7 mm, a reduction of 35-50%. The optimized method (Method 3) reduced mean errors in 64-68% of patients across all directions, compared to 68-71% for Method 2 relative to Method 1. Notably, Method 3 reduced the incidence of gross errors by 67% compared to no averaging. Conclusion: The novel optimized averaging method for EPID-based setup corrections significantly improves setup accuracy in head and neck radiotherapy, outperforming conventional techniques. This approach substantially reduces CTV-PTV margins and the occurrence of gross errors, potentially allowing for more conformal dose distributions and reduced toxicity. These findings have important implications for improving treatment precision and outcomes in head and neck cancer radiotherapy. © 2026 Mashhad University of Medical Sciences. All rights reserved.