Comparing two different noise magnitude estimation methods in CT using Virtual Imaging Trials

Noise magnitude is a key indicator of CT image quality. As such, different methods have been implemented to measure noise in vivo. The two most common approaches consider the estimation of noise in soft tissues (GNI) and in the air surrounding the patient (A/Rn). An effective comparison of the two methods is practically impossible because the anatomy of the patients, the scanner models, and the acquisition protocols introduce nonsystematic biases. An objective and comprehensive analysis can be performed only by repeated scans of the same patients which is unethically impractical due to the excessive radiation burden. A feasible solution is to simulate these repeated scans using Virtual Imaging Trials (VITs). In this study, XCAT phantoms were imaged 50 times at three dose levels using a VIT platform (DukeSim) and reconstructed with three kernels using both FBP and IR algorithms for Chest and Abdominopelvic protocols (a total of 1800 image datasets). By applying different HU thresholds, A/Rn was calculated and compared with GNI. The differences between A/Rn and GNI ranged between -64% and -47% in abdomen and between -57% and -33% in chest studies. The A/Rn's underestimation of noise did not show any correlation with kernels, reconstruction algorithms, and dose levels. Noise measured in the air surrounding the patient cannot represent the noise magnitude in soft tissues. Care should be exercised when designing and implementing optimization actions relying on methods that estimate image quality in areas of the diagnostic images that are not clinically relevant.