Response surface methodology for predicting optimal conditions in very low-dose chest CT imaging

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• CT protocol optimization is complex due to variations in reconstruction algorithms and automated exposure control systems.
• Standardizing dose reduction techniques across different manufacturers and scanner models remains a challenge.

Purpose of the Study:

• To investigate the feasibility of using the design of experiments (DOE) for optimizing computed tomography (CT) protocols.
• To identify optimal CT parameters for dose reduction while maintaining diagnostic image quality in chest examinations.

Main Methods:

• A Doehlert matrix was employed to design experiments on a 128-slice CT scanner using an anthropomorphic chest phantom.
• Lesion detectability was evaluated using model and human observers with iterative (ASIR-V) and deep learning-based reconstruction (DLIR-L, DLIR-H).
• Second-order polynomial functions modeled the effects of noise index (NI) and ASIR-V percentage on dose and observer performance.

Main Results:

• Optimal conditions predicted were NI = 64, 60% ASIR-V, and DLIR-H reconstruction, showing good agreement with human observer results.
• The study indicated a potential 64% dose reduction using DLIR-H compared to 60% ASIR-V, without compromising lesion detection.
• Bland-Altman analysis confirmed strong agreement between predicted optimal conditions and experimental human observer findings (mean absolute difference -0.01 ± 3.16).

Conclusions:

• The proposed DOE method effectively predicts optimal conditions for low-dose chest CT.
• This approach minimizes the number of experiments required for protocol optimization.
• Diagnostic image quality can be ensured while significantly reducing radiation dose.