WE-D-18A-07: Development and Evaluation of Automated Noise Measurement Technique in CT Images

Authors


Abstract

Purpose:

To develop and evaluate an automated technique which measures the noise level in CT images.

Methods:

Candidate regions of interests (ROIs) with circular shape were selected with a criteria where the pixel intensities on Gaussian filtered image lie within a predefined range. The ROIs were further refined with the sum of absolute gradients within each candidate ROI being used as a ROI feature to distinguish homogeneous ROIs. A histogram of ROI feature was created, and the ROIs with feature values less than the mode of histogram were selected as homogeneous ROIs. Connected component analysis was then used to remove the non-clustered ROIs to restrict the location of ROI in desired organ. Standard deviation of pixel intensities was calculated for each cluster of homogeneous ROIs to represent the local noise level, and its mode value was used to represent the noise level. Fifty abdomen CT images were collected, and five observers participated to manually measure the noise level with a guidance to draw same size ROIs. Manual results were then averaged to establish a reference value, and a correlation coefficient(r) between a reference value and our automatic result. Intra- and inter-rater agreement were compared to demonstrate the performance of our proposed technique.

Results:

We confirmed successful operation of automatic homogeneous region localization and noise level measurement. Intra- and inter-rater agreement(r) ranged from 0.70 to 0.95, and from 0.79 to 0.94, respectively. Agreements between reference and manual measurements by raters ranged from 0.85 to 0.95, and those between a reference and an automated result by our proposed method was 0.90.

Conclusion:

Our developed technique has successfully performed the noise measurement task in all the test images by showing an agreement of 0.90 with the reference measurement. Our proposed method enables the automated assessment of CT image noise.

Ancillary