Abstract Purpose: This study aimed to compare human observer detectability (d′) obtained from receiver operating characteristic (ROC) and two-alternative forced choice (2-AFC) experiments using synthetic images. Methods: Synthetic images containing three signal profiles (flat, Gaussian, and designed) were generated at two tube current settings of 100 mA and 200 mA. Image synthesis was performed using a hybrid approach that combines a computational phantom with real CT images to obtain system characteristics, including the task transfer function (TTF) and noise power spectrum (NPS). Human observer experiments were conducted using ROC rating and the 2-AFC paradigm under matched viewing conditions. The d′ was calculated for each condition. Pearson correlation analysis was performed to assess the relationship between ROC and 2-AFC detectability. Results: d′ values increased with increasing tube current for all signal profiles. The 2-AFC paradigm consistently produced higher d′ values than the ROC method under all conditions. A strong positive correlation between ROC and 2-AFC (r = 0.93), that the observed correlation is statistically significant (p < 0.01), indicating consistent detectability trends across methods. Conclusion: ROC and 2-AFC experiments on synthetic images demonstrate strongly correlated detectability trends, although 2-AFC systematically yields higher d′ values. These findings support the use of synthetic images for observer studies and provide a practical guidance to evaluate selecting appropriate experimental paradigms for imaging performance.
Putri et al. (Thu,) studied this question.