Evaluation of statistical low-contrast detectability (SLCD) with various confidence levels in CT images and their...

This study aims to evaluate statistical low-contrast detectability (SLCD) at multiple confidence levels (CLs) in computed tomography (CT) images and to determine the smallest detectable object sizes compared with human observers (HO).

SLCD was calculated on a homogeneous region of the CT images using grids of varying cell sizes. For each grid, mean CT numbers were calculated and the standard deviation was used to derive the minimum detectable contrast (MDC) as a function of CL via the separation distance between object and background distributions. CLs from 95% to 99.99% were evaluated. The target sizes (smallest object sizes detected) were compared with visual assessment performed by 30 experienced medical physicists. The evaluation was carried out on the low-contrast module (module #2) of the ACR CT phantom scanned by GE Revolution EVO 128 slice CT scanner in various tube currents (80-200 mA), tube voltage (80-140 kV), and field of view (170-230 cm).

At 95% CL, SLCD estimated target sizes were approximately half of those identified by HO. Increasing CL produced larger SLCD target sizes, where 99.99% CL yielded target sizes comparable to HO results across all parameter variations.

SLCD outcomes depend strongly on the selected CL. To align with human performance at 6 HU, using ∼99.99% CL is recommended. This has implications for quality control (QC) and protocol optimization where statistical methods substitute for human reads.