Phantom study
Our study involved the CT Electron Density Phantom (CIRS,
Norfolk,
model 062).
This phantom consists in two body parts made of soft tissue equivalent epoxy resin.
The use of inserts with known elemental composition and Relative Electron Density (RED) indicated by the phantom’s manufacturer allowed a conversion of the RED value to the associated CT attenuation values of each pixel [14].
The following RED values were included in the phantom: 0.190,
0.489,
0.949,
0.976,
1.043,
1.052,
1.117 and 1.512.
The RED values equal to 0.949,
1.043,
1.052 and 1.117 were defined as fat,
muscle,
liver and aorta substitutes respectively. A special insert dedicated to the use of a syringe filled of regular water was used. The phantom was scanned with inserts in identical locations between acquisitions.
The same phantom and inserts were used for all image acquisitions in order to avoid the variation of the RED values for the tissue substitute materials among several phantoms [15].
CT images acquisition
- CT scans were acquired on two identical 64-section multi-detector row helical CT scanners (LightSpeedTMVCT,
General Electric Healthcare),
with the following parameters: 120 kV,
AutomA,
automatic exposure time,
0.5 s gantry rotation time,
500 mm field of view,
slice thickness and slice distance of 0.625 mm,
pitch equal to 0.984,
standard filter,
noise index was 23,
Filtered Back Projection reconstruction.
This protocol for image acquisition suited to liver diagnosis was defined to meet the radiologists on-going needs at the beginning of the study.
It had been set while installing CT scanners,
and kept identical for quality control throughout the time of the study regardless of any changes came from radiation dose optimization or software update.
- CT scanners were calibrated according to the manufacturer’s specifications during maintenance operations.
- Twenty-two total phantom scanning sessions were performed just after maintenance within five and half years on each CT equipment.
- Phantom images were imported into a dedicated image processing software (ArtiscanCT software,
version 4.0.1,
Aquilab,
France).
Image quality analysis
Regions of interest (ROIs) were automatically extracted by the software on one axial slice chosen at the mid-z-axis thickness of the phantom to ensure image uniformity and avoid partial volume effects.
The mean Hounsfield numbers (HU) and standard deviations (SD) were determined in circular ROIs (diameter 10mm) with the center coinciding with the center of inserts.
Standard deviations were defined as tissue noise.
Therefore,
the tissue-to-noise ratio was defined as the ratio between the average and the standard deviation of pixel values extracted from a ROI inside tissue inserts.
Statistical analysis
First,
the CT attenuation values and tissue-to noise ratios were calculated for each RED value and were expressed in terms of mean,
standard deviation of the mean,
minimum,
maximum for each CT scanner for five and half years.
For each RED value and for water,
the HU values were compared between the two CT scanners by using a two-sided paired Student’s t test.
If no significant difference was observed,
measurements were considered reproducible (P>0.05) (Table 1 and 2).
Second,
in order to assess the agreement between attenuation values and tissue-to-noise ratios from the two CT equipments,
a Bland and Altman (B&A) analysis was performed [16,
17].
The regression line of the differences was drawn in order to help in detecting a proportional difference (Figure 2 and 3).
Analysis were performed in Microsoft Excel software (Microsoft Corp.,
Redmond,
WA USA).