Abstract Background A contrast detail (C‐D) curve is one important tool for computed tomography (CT) dose optimization and it can be developed using detectability index (d′). This approach involves constructing a d′ map over a series of task objects with varying sizes and contrasts. However, generating a smoothed d' map can be computationally intensive. Purpose This study aims to develop and implement a smooth d′‐based C‐D curve for optimizing CT liver imaging protocol at varying tube voltages. Methods The task transfer function (TTF) and noise power spectrum (NPS) were derived from CT images of the ACR 464 phantom acquired at 80, 100, 120, and 140 kVp. A matrix of task objects was defined, incorporating both Gaussian and Designer signal profiles, with object sizes ranging from 1 to 15 mm and contrast levels from −1 to −15 HU. To simulate clinical liver imaging, each task object was modelled on a 50 HU background. The d′ values were calculated for each task using two model observers: non‐prewhitening (NPW) and NPW with eye filter (NPWE). A d′ map was generated by iterating these measurements across all tasks, and subsequently interpolated to produce a smooth d′ map. Finally, a smooth C‐D curve was obtained by applying a contour‐finding algorithm on the interpolated d′ map using a specific d′ cut off. The results of C‐D curves at d′ = 3 were compared with those from human observers (HOs). Results Smooth C‐D curves and their corresponding synthetic image matrices were successfully generated for the image data. Increasing the tube voltage leads to an improved object detectability limit for the simulated liver examination. This trend was consistently observed for both Gaussian and Designer signals as evaluated by the NPW and NPWE models. In comparison, the NPW model yielded higher detectability estimates than NPWE. Comparison with HO data suggests that NPWE produce more similar detection patterns with human observation. Conclusion The smooth d′‐based C‐D curve was successfully developed and implemented as a robust tool for CT protocol optimization.
Anam et al. (Sun,) studied this question.