ESTRO 2025 - Abstract Book

S2974

Physics - Image acquisition and processing

ESTRO 2025

1 Medical Physics, Memorial Sloan Kettering Cancer Center, New York, USA. 2 Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, USA Purpose/Objective: To devise a predictive noise assessment model to compare the essential noise difference of CT images scanned between scanners of distinct models and manufacturers. Material/Methods: An automated assessment of the global noise index (GNI) was used to measure noise levels in clinical CT images, with patient size determined by the water equivalent diameter (WED). A total of 1581 examinations were conducted on five scanners: three from vendor P and two (a PET/CT-S p and a CT-S c ) from vendor S, classified by anatomical regions (brain, head and neck (H&N), breast, lung, abdomen, pelvis, spine). Protocols were compared using GNI as the dependent variable at two levels: (1) between different models from the same manufacturer and (2) among various manufacturers. Statistical analyses included Student t-tests for level (1) and one-way ANOVA for level (2), with p < 0.05 considered significant. Given that quantum noise is inversely proportional to the square root of the signal received by the detector, a noise predictive model was developed through regression analysis of GNI in relation to WED, represented by the equation: GNI = A*sqrt(exp(u*WED))+B, where u is the attenuation coefficient of water at 120 kV. Estimated GNIs for reference WEDs were derived from this model for protocol comparisons across different scanners. Results: The average GNI (GNI m ) showed no significant variation among the H&N, lung, and pelvis subgroups when comparing S p and S c from the same manufacturer. ANCOVA analysis with WED as a covariate revealed no significant differences in the abdomen and spine subgroups. A one-way ANOVA indicated that, except for the H&N subgroup, GNI m values in the S examinations differed from those obtained using the P scanners