Click here to

Session: Imaging: Dose and Image Quality in CT and X-ray Imaging [Return to Session]

Local Noise and Size of Region of Interest in CT Images

G Li1*, Y Liang2, (1) University of Maryland, Baltimore, Baltimore, MD, (2) ,


WE-IePD-TRACK 2-3 (Wednesday, 7/28/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]

Purpose: This study aims to investigate the correlation between the automated local image noise estimates with the manual noise estimates and effects of the size of region of interest (ROI) on the correlation.

Methods: 76 clinical non-enhanced axial abdominal scans are randomly selected among the exams performed using the same protocol on the same scanner in the same quarter. One slice of each of 76 scans locating middle liver is used for comparison. For manual method, the standard deviation (SD) Hounsfield Unit (HU) within a 35-pixel radius circular ROI placed in a relatively uniformed area in the liver in ImageJ is used as the noise estimation. For local noise analysis, the SD HU of the n x n ROI (n=3, 7, …., 67) centering each body pixel is used as the noise estimate around that center pixel (Fig. 1). The local noise of each and every body pixel is calculated. This approach is automated in Matlab. The noise of the automated method is then compared with that of the manual method using linear regression approach.

Results: The lower 25-percentile noise of all body pixel in general represents more uniform area shown in (Fig 2). Although the noises of the manual and automated method are well correlated, the slope and variance of residuals (Fig 3) increase with the square ROI size, suggesting that selecting the same smaller ROI for noise comparison across patients and scanners is required.

Conclusion: The noise in clinic images estimated by a manual ROI is linearly correlated with the mean of lower 25 percentile of local noise obtained by an automatic method and a fixed small ROI is critical for intra- and inter-patient and scanner noise performance comparisons.



    Not Applicable / None Entered.


    IM- CT: Quality Control and Image Quality Assessment

    Contact Email