Insights, robustness and practical considerations of global noise level measurement in chest CT

Accuracy of global noise level (GNL) has been benchmarked predominantly for soft tissue. This work extends the use of GNL to lung tissue and investigates the effect of patient, protocol and method parameters on the robustness of the algorithm.

Optimal GNL algorithm parameters for lung tissue were established after comparing manual measurements in one hundred patient images. Automatic determination of patient-specific GNL was performed from all chest CT slices with outlier verifications and lung tissue area estimation. Accuracy of GNL mean and variation were assessed with increasing number of included slices to convert GNLs into a stack-overarching parameter. Finally, GNL in soft and lung tissue were compared in identical patient images.

Optimal GNL parameters for lung tissue were: 4 mm x 4 mm sliding window, Hounsfield units below -590 HU, and 1 HU histogram bin width. In chest CT slices with a lung tissue area above 40 cm², 99.8 % of GNL calculations were clinically relevant. A 95 % accurate mean lung GNL required minimally 17 equidistant slices. Finally, lung GNL in thicker slices was higher than soft tissue GNL, but the reverse was observed in thin slices.

This study provides insight into important practical considerations on the calculation and subsequent use of GNL for monitoring and image quality evaluation.