This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Radiation physics, Radioprotection / Radiation dose, Computer applications, CT, Image manipulation / Reconstruction, Computer Applications-General, Physics, Image verification
Authors:
H. Pasquier1, F. Gardavaud2, A. Rahmouni1, A. Luciani1; 1Creteil/FR, 2Paris/FR
DOI:
10.1594/ecr2017/C-3121
Methods and materials
An automated tool was developed to extract noise texture and magnitude metrics from 10 chest and abdomen clinical CT-scans (Discovery CT750 HD; GE Healthcare,
Wisconsin) performed for oncology follow-up by measuring a global Noise Power Spectrum (NPS).
First,
a noise image was obtained by subtracting adjacent axial slices ( Fig. 1 ) from one another to remove the main anatomical structures that tended to be correlated between images (Fig. 2).
Then,
a segmentation algorithm was applied to pinpoint edges of remaining anatomical structures ( Fig. 3 ).
Finally,
an algorithm was developed to loop through the image and extract all Regions Of Interest available; i.e.
within the body region and not containing edge pixels.
ROI size was set to 32x32 pixels allowing sufficientpixel sampling while minimizing the effects of CT numbers nonuniformities.
Overlapping of ROIs was not considered.
The global NPS was computed as the average NPS within all the ROIs extracted.
Noise magnitude (i.e.
area under the 1D-NPS curve) and noise texture (i.e.
peak frequency,
noted fpeak) were measured for chest and abdomen on image subsets of all CT scans reconstructed using Filtered Back Projection (FBP),
Adaptive Statistical Iterative Reconstruction 30 % (ASIR30),
50 % (ASIR50) and Model-Based Iterative Reconstruction (MBIR).
Noise magnitude and noise texture were compared for each reconstruction of abdomen and chest CT images using a Kruskall-Wallis one-way ANalyse Of Variance (ANOVA) with Dunn post hoc test; P<0.05 was considered as statistically significant.