SU‐F‐I‐44: Using the Air Signal to Characterize Image Noise in CT

Abstract

Purpose:With increasing emphasis on CT image quality and low dose, quality activities analyzing noise levels in images will need to be increasingly automated. This work describes a new automated method for characterizing noise in CT. To obviate the need to segment out anatomical structure, this work introduces the use of noise statistics of air to understand the noise in the scanned object.Methods:Scans of a uniform water phantom were acquired at four dose levels. Using a modified version of the global noise index, the image noise was calculated in both the air surrounding the phantom and in the phantom itself. The method involved subtracting adjacent image slices, breaking up the image into 7 mm square regions of interest (ROI), thresholding the ROIs by CT number to only include ROIs of air or water, and then taking the mean of the air or water ROIs’ standard deviation values. We refer to this metric as the meanGNI.Results:Not only does the meanGNI within the phantom decrease as dose increases, but also the meanGNI in the air exhibits the same behavior.Conclusion:The meanGNI in air is a useful proxy for meanGNI or GNI in the scanned object. Our air based method does not involve any edge detection or advanced signal processing. Therefore, no tuning of segmentation parameters is needed to be applicable to different types of anatomy (e.g. brain vs. lung).Research and equipment support provided by GE Healthcare.