After obtaining IRB approval,
we retrospectively evaluated image quality and radiation dose in 36 patients scanned using the prototype automatic kV selection tool.
The scans were clinically indicated and informed consent was not required. Results were compared to results from control patients scanned using identical protocols,
phase of enhancement,
slice thickness,
oral contrast,
and patient width (within 5 cm,
measured at the level of the liver dome). Whenever possible,
scanner type was also matched. Auto kV patients were scanned on a 128-slice dual-source scanner (Definition FLASH,
Siemens Medical Solutions,
Forchheim,
Germany),
which was provided by the manufacturer,
via a grant to Mayo Clinic.
Prototype Automatic kV Selection Tool
The automatic kV selection tool uses the CT topogram,
the corresponding attenuation information from the patient,
and the user-specified exam type to determine the optimal kV to achieve the best dose efficiency for the entire length of the scan.
Once the volume to be imaged is prescribed,
patient- and task-specific mAs curves are calculated for all kV levels (80,
100,
120,
and 140 kV) based on the given scan range,
patient anatomy,
and user-selected contrast behavior (i.e.,
identification of scan type,
such as routine abdominal CT versus abdominal CT angiogram) necessary to deliver the desired image quality.
For each diagnostic task,
a contrast gain setting that described the dose reduction and acceptable noise level was selected (very strong,
strong,
average,
weak,
very weak). This contrast gain setting was selected by CT physicists in our practice,
who measured acceptable noise levels from routine exams. All contrast-enhanced abdominopelvic CT exams were performed using the very strong or strong contrast gain settings. According to the contrast gain setting and the automatic exposure control software (CAREDose4D,
Siemens Healthcare),
a tube current modulation curve over the whole scan range for each kV was calculated.
The radiation dose output (CTDIvol) was then estimated based on these kV-specific mAs curves for all of the kV levels.
The Auto kV tool then checked the system to determine which kV was the most dose-efficient and whether or not the optimal setting was possible (i.e.,
due to tube current limits,
pitch settings,
scan range,
etc).
If this setting was not possible,
the next best kV setting was suggested.
The tool would then display for each tube potential,
the estimated dose reduction,
the new quality reference mAs (that is designed to provide consistent contrast-to-noise ratio according to the contrast gain setting),
and pitch. Other than kV,
mAs and pitch,
all other acquisition and reconstruction parameters remained identical to the base scanning protocol (at 120 kV).
For each patient that underwent contrast-enhanced abdominopelvic CT with the prototype automatic kV selection tool (Auto kV),
the CT technologist performed the topogram and prescribed the volume to be imaged with the appropriate task-specific CT protocol,
recording the estimated CTDIvol if the patient would have been scanned using the base scanning protocol at 120 kV. The Auto kV selection tool was then activated by the CT technologist without changing the volume to be imaged. The technologist was instructed to select the kV with the greatest predicted radiation dose reduction that the system could deliver,
and record the new estimated CTDIvol.
After performing the exam,
the delivered CTDIvol was recorded as well.
Radiation Dose Reduction Calculation
Radiation dose reduction gained by using the prototype Auto kV tool was estimated using two methods. In the primary method for estimating dose reduction,
each patient served as his/her own control. In this calculation,
the estimated CTDIvol (mGy) that would be required to scan the imaged volume using the base protocol at 120 kV (e.g.,
CT enterography,
CT urogram,
biphase liver) was compared to the CTDIvol that was recorded at the completion of each examination.
The secondary method matched each patient undergoing CT imaging with the Auto kV tool to a control patient undergoing an identical type of CT exam. Each Auto kV case was matched to a control so that the control exam had the same slice thickness,
phase of enhancement,
oral contrast,
and patient width (within 5 cm).
Width was measured from skin to skin at the level of the superior aspect of the liver on the coronal topogram. Delivered radiation dose in CTDIvol in control patients was extracted from the medical record. CT examinations in these control patients also served as the reference for expected image quality at 120 kV given the mix of diagnostic tasks and patient sizes in our cohort.
Image Analysis
A qualitative and quantitative image analysis was performed. The qualitative analysis was performed by two gastrointestinal radiologists with 12 and 14 years of experience,
respectively,
using a modified European Quality Criteria scoring system [14-16].
The two readers evaluated image quality metrics on all of the selected CT examinations in randomized fashion,
blinded to the use of the kV selection tool and all other acquisition parameters,
including radiation dose. For image quality analysis,
readers examined image sharpness,
noise,
noise texture and diagnostic confidence. Radiologists rated image sharpness as the critical reproduction of sharp structures such as the portal veins,
bile ducts,
mesenteric vessels,
and renal arteries from 1-3 (1 = very sharp; 2 = questionable blurriness; 3 = noticeable blur or slice thickening). Noise was also rated along a 3-point scale (1 = less noise than usual; 2a = optimal noise similar to perceived routine noise; 2b= optimal noise,
considered slightly more noise than routine but also adequate; 3 = noise affects interpretation). Noise texture was rated along a 4-point scale (0 = no noticeable change; 1 = after altering CT window settings,
no noticeable change in noise texture; 2 = perceptible change in noise texture; 3 = change affects confidence or blotchiness). Diagnostic confidence was rated along a 4-point scale (1 = fully confident; 2 = probably confident; 3 = confident under limited conditions; 4 = unacceptable).
For quantitative analysis,
CT number measurements were made within the aorta,
right lobe of the liver (excluding vessels,
at the level of the bifurcation of the portal vein),
and portal vein using a region of interest. Image noise was measured within the subcutaneous fat (two measurements in the anterior and posterior subcutaneous fat,
respectively). Iodine contrast-to-noise values were reported for the aorta,
portal vein,
and liver parenchyma using the CT number within these structures divided by the mean noise in the subcutaneous fat.
Statistical Analysis
The analysis was conducted in two phases.
The first phase focused on the within-patient changes in the patients scanned with the Auto kV tool. The paired Wilcoxon signed-rank test was used to examine the change between the CTDIvol after scanning with the Auto kV tool compared to the CTDIvol predicted using the base protocol 120 kV prescription. The second phase compared the CTDIvol from Auto kV exams to that in the matched control exams using the 120 kV base protocol without Auto kV and also utilized paired Wilcoxon signed-rank test.
For comparisons of image quality analysis by each radiologist reader for Auto kV and control CT examinations,
frequencies were examined for each reader for each image quality metric. For quantitative analysis,
CT number in the aorta,
liver and portal vein,
as well as iodine contrast-to-noise ratio for these structures,
were compared between Auto kV patients and controls using the Sign test.
Spearman correlations were used to test for association of patient width,
dose reduction and percent dose reduction. Patient width was then used to categorize the subject population of cases and controls. Patients were grouped into sizes based on size-specific recommendations for lower tube potential scanning for contrast-enhanced abdominopelvic CT (i.e.,
small patients = lateral width < 36 cm; medium-sized patients = 36-41 cm; large patients < 41 cm in lateral width) [17]. Kruskal-Wallis tests were used to test for a relationship of patient size with dose reduction and percent dose reduction. Percentage dose reduction was analyzed according to patient size as well as according to kV selected by the Auto kV tool,
and is shown descriptively and with analysis outcomes in box-whisker plots.
Statistical analyses were conducted using SAS (version 9.22; Cary,
NC).
No correction for multiple testing has been applied to reported p-values.