Effect of CT noise on detectability of test objects

Three types of test objects were superimposed on noisy backgrounds and observed by 58 subjects: large low-contrast disks to simulate tumors, small disks to simulate calcifications, and bars to simulate blood vessels. Three types of noise were used: random noise, CT noise (reconstructed random noise), and an actual water scan taken with an EMI CT1010 scanner. In addition, each image was smoothed by a factor of two, giving six test patterns altogether. Photographic images were presented to 25 lay observers, while 25 other lay observers and eight radiologists used a viewing console. The photographs produced a slightly lower true positive rate than the console, although the degree of confidence in the identifications was slightly greater. The anticorrelation property of CT noise gives it a "lumpy" structure, which aided the identification of bars but impeded the identification of small disks. Smoothing did not significantly affect the detectability with random noise, but with CT noise, smoothing improved the detectability of bars and large disks.