ESTRO 2025 - Abstract Book

S3623

Physics - Quality assurance and auditing

ESTRO 2025

1 Medical Physics Unit, Centre Marie Curie, Valence, France. 2 Medical Physics Unit, CHR Metz-Thionville, Metz, France. 3 Radiation Oncology Unit, Centre Marie Curie, Valence, France. 4 Centre de Recherche en Automatique de Nancy, CNRS, Nancy, France Purpose/Objective: The gamma index, introduced by Low et al. (1998), is a cornerstone of radiotherapy quality assurance (QA), comparing calculated and measured dose distributions. However, it lacks sensitivity to subtle but clinically significant discrepancies and cannot differentiate between positive and negative deviations. This study introduces a novel algorithm derived from the gamma index, integrating signed data analysis, 3D surface visualizations, inverse gamma passing rate isolines, and a new metric termed the "D-map."

Material/Methods: The algorithm, developed in Python (see Figure 1), features: •

Closest Point Identification: For search radii (Ri), the nearest points are determined based on dose and distance criteria. • Dose Difference Recording: Dose discrepancies at these points are used to construct cumulative 3D histograms, with optional signed analysis, to visualize passing rates via isolines. • D-map Construction: The D-map quantifies the norm of vectors minimizing dose and distance deviations, incorporating signed information.

Figure 1-Step-by-Step Process for Advanced Gamma Index and D-map Calculation

The algorithm was evaluated in two scenarios: 1. Patient QA : Comparison of dose distributions with a simulated 3 mm shift. 2. Machine modeling : Comparison of a measured FourL field on a PTW matrix with two multi-leaf collimator models from the RayStation TPS. Results: In a patient QA example, passing rates of 95% were achieved with dose criteria of 2.1%, 0.6%, 0.4%, 0.3%, and 0.2% for radii 1–5 mm, respectively. Allowing negative deviations outside the target volume reduced dose criteria to 1.6%, 0.5%, 0.3%, 0.2%, and 0.2%. For the FourL analysis, Model 1 achieved tighter dose criteria for the 95% isoline compared to Model 2, alongside larger passing rate surfaces under the 3D curve and lower D-map values, indicating superior accuracy in dose modeling (Figure 2).