Outline of the multi-phase data-averaging technique (Fig.
multiple data acquisitions of an object and their averaging at CT yields lower image noise than dose single data acquisition .
Our new method uses three different cardiac phase images to reduce image noise. While coronary CTA with prospective ECG-triggering is usually performed in the center of the mid-diastole phase,
data can be acquired at the off-center of the temporal window if redundant scan time (padding) is added.
We set the range and center of the phase window at 70-80% and 75% of the R-R interval,
We first generated three sets of volume data at 70-, 75-,
and 80% of the R-R interval. Next,
as images at each phase were spatially misaligned, we applied non-rigid registration to align the 70- and 80% images to the 75% image. We performed image registration with the Insight Segmentation and Registration Toolkit (ITK; http://www.itk.org) and applied the ITK particle analysis tool(BSplineDeformableRegistration module).
This registration module includes mutual information metric and deformation field-based on the 3rd-order B-Spline function.
It can perform non-rigid deformation that includes translation,
After running the registration module,
the 70- and 80% images were aligned to the shape of the 75% images on a pixel-by-pixel basis.
In the last step we performed weighted averaging of the three images and generated a de-noised image. Linearly weighted averaging was with ImageJ software (National Institutes of Health,
The weighted factor of the 75% image was 0.4; it was 0.3 for the aligned 70% and 80% images.
The three multiplied images were summed using the ImageJ particle analysis tool (ImageCalculator plugIn),
and a de-noised image was generated (Fig.
Fig. 1: Outline of the multi-phase data-averaging technique. At first, three sets of consecutive volume data at the 70-, 75-, and 80% phase of the R-R interval were prepared. Next, the 70- and 80% images were aligned by non-rigid registration to the 75% image. Finally, weighted averaging of the three images was performed and a de-noised image generated.
Fig. 2: A 74-year-old man with a body mass index of 26.9 kg/m2. A sample transverse CT image of the ascending aorta and left main coronary artery (120 kV, 720 mA) is shown. (a) Conventional 75% image. The ROI shows CT attenuation of 372.0 HU and image noise of 24.8 HU. (b) De-noised image. The ROI shows CT attenuation of 369.2 HU and image noise of 19.2 HU.
Institutional review board approval was obtained for this retrospective study; informed consent was waived.
We enrolled 30 patients (25 men,
median age 68 years,
35 - 80 years) who underwent coronary CTA with prospective ECG-triggering.
If the patient's resting heart rate exceeded 65 beats per minute (bpm),
we orally administered 20-40 mg of metoprolol (Selokeen) 60 min before CT examination.
All had normal renal function (serum creatinine level < 1.5 mg/dL),
no history of allergy to contrast agents,
and a left ventricular ejection fraction of more than 40%.
All coronary CTA examinations were performed using a 64-slice CT scanner (LightSpeed VCT XT scanner,
GE Healthcare) with prospective ECG triggering (SnapShot Pulse).
The volume of the contrast material was adapted to the patient’s body weight; all patients received 0.6 ml/kg of nonionic contrast material (Iomeprol,
Iomeron 350 mgI/ml) injected at a fixed duration of 10 sec followed by 20 ml of a 0.9% saline solution injected at the same flow rate as the contrast material.
The scanning parameters were as follows: detector configuration of 64 × 0.625 mm,
gantry rotation time of 350 msec,
tube voltage of 120 kV,
and tube currentof 400-750mA. In all patients the phase window during which the patient was exposed was limited to 70-80% of the R-R interval. Axial images were reconstructed with a slice thickness and a reconstruction interval of 0.625 mm using a medium soft-tissue convolution kernel (standard).
We prepared three sets of consecutive volume data at 70-, 75-,
and 80% of the R-R interval and performed multi-phase data-averaging (Fig.
1).The effective radiation dose of CTA was calculated as the product of the dose-length product (DLP) multiplied by a conversion coefficient for the chest (k = 0.017 mSv/mGycm) .
One board-certified radiologist with 8 years of experience in cardiovascular radiology calculated the contrast-to-noise ratio (CNR) in the proximal right (RCA) and left main (LMA) coronary arteries using these steps: First,
attenuation in a region of interest (ROI) in the proximal RCA and the LMA was measured.
The vessel contrast was calculated as the difference in mean attenuation between the contrast-enhanced vessel lumen and the adjacent perivascular tissue.
image noise was determined as the standard deviation (SD) of the attenuation value in an ROI placed in the ascending aorta.
The CNR was calculated as the ratio of vessel contrast over noise.
Two board-certified radiologists (with 8 and 17 years of experience in cardiac radiology,
respectively) assessed overall image quality for each major branch of the coronary arteries (right coronary-,
left anterior descending coronary-,
and left circumflex coronary artery [RCA,
If their data analysis disagreed a final decision was reached by consensus.
The overall image quality of each coronary artery segment was rated on a 5-point score where 5 = excellent (no motion artifacts or noise-related blurring and excellent vessel opacification),
4 = good (minor motion artifacts or noise-related blurring,
good vessel opacification); 3 = acceptable (some motion artifacts or noise-related blurring,
fair vessel opacification),
2 = suboptimal (marked motion artifact or noise-related blurring,
poor vessel opacification),
and 1 = nondiagnostic.
Images with a score of 3 or higher were considered diagnostic.
We compared conventional 75%- and de-noised images for differences in our image quality parameters (image noise,
CNR) using the paired t-test.
The image quality score of the coronary arteries on the two image data sets was compared with the Wilcoxon signed-rank test.
We looked for a linear relationship between the noise reduction rate (%) and the time (msec) for the R-R interval during the acquisition of the CT scans using the Pearson correlation coefficient.
Differences were considered to be statistically significant at p < 0.05.