Literature Review on the Image Filtering Approach
Review the following journal article. The article has been uploaded to WebCT.
Szucs-Farkas Z, et al. Nonlinear Three-dimensional Noise Filter with Low-Dose CT Angiography: Effect on the Detection of Small High-Contrast Objects in a Phantom Model. Radiology 2011;258:261-269.
The literature review should include the answers of the following questions.
- Background of the study (What have the other studies already done?)
In the past decade, multidetector computed tomographic (CT) angiography has become a major tool in the noninvasive evaluation of the aortoiliac system. The use of ionizing radiation is a well-known drawback to the method. Frequent follow-up examination in patients with aortic dissection or after endovascular aortic aneurysm repair (EVAR) can lead to a substantial cumulative dose. Among the numerous possibilities, the use of reduced CT tube energy is a simply way to reduce patient dose at multidetector CT angiography. Unfortunately the low-kilovoltage CT images are noisier than the normal-dose counterparts, which can hamper acceptance for diagnostic purposes.
- Rationale of the study (What have not been done by other studies?)
Image post processing by using noise filters, built-in soft reconstruction kernels, or adaptive statistical iterative reconstruction algorithms can efficiently reduce noise. Furthermore, before adaptive statistical iterative reconstruction became available in the latest-generation of CT scanners, it was a very time-consuming technique. To date, the effect of noise reduction on image quality is well documents however the results on diagnostic accuracy are limited. The research team is trying to show the statistically strong evident on accuracy for the post processing clinical images.
- Research question of the study
The purpose of the investigation was to determine the effect of a nonlinear, three-dimensional optimized reconstruction algorithm (3D-ORA) on the detection rate of simulated endoleaks at multidetector CT angiography by using 80- and 100- kVp tube energies in an abdominal aortic aneurysm phantom.
- Data analysis/visualization techniques used in the study
The Phantom:
Stationary conditions in an abdominal aortic aneurysm treated with EVAR in intermediate and thick phantoms were simulated by merging a cylindrical phantom into one of two water-filled containers with diameters of 30-40 cm, which were used to simulate intermediate-sized and large patients, respectively. All phantom components was based on measurements in 10 consecutive patients after EVAR with type-II or III endoleaks at 100 kVp multidetector CT angiography. The phantom consisted of 18 epoxy resin disks, 80mm in diameter, containing a central hole measuring 20 mm for the simulated blood-filled aortic lumen.
CT Data capturing:
CT data were acquired with a 64-detecor row CT scanner by using the routine protocol for the thoracoabdominal multidetector CT angiography. The tube potential was set at 80, 100 and 120 kVp for the 30-cm phantom and at 100 and 120 kVp for the 40-cm phantom. The built-in automatic tube current modulation software was switched on during the data acquisition. The quality reference tube current-time product was set at 260 mAs for the 80kVp tube voltage, and 160 mAs for the 100- and 120- kVp tube voltage, corresponding to the routine protocol. All images were transferred and stored in the PACS. The volume CT dose index provided by the CT scanner was recorded for each series.
Noise Filter and Image Post-processing:
The prototype nonlinear, 3D ORA filter generalizes the 2D smoothing technique on all 3D of space and come with the edge detection algorithm. The filtered images are built from a mix of local weights of filtered data in the 3D and the original data, whereas the filter effect is reduced in regions with a contrast change.
In a preliminary series, the 100-kVp data sets with both phantom sizes had been post processed with five different filter presets, and the images were subjectively assessed for noise and sharpness of endoleaks by the study investigators.
Analysis of Objective Image Quality:
Attenuation of the phantom components was measured by the study investigator in all series. The circular regions of interest were drawn as large as possible, carefully avoiding the edges of the component; the region-of-interest size averaged 1043 mm square with 17mm square in the water container. Noise was defined as the standard deviation of attenuation in the simulated thrombus. All measurements were performed 10 times in different sections to avoid inaccuracies, and the mean was used for calculating the CNR for the aorta as follows:
CNR = (HU a0 – HU Thr) / noise ;
Where HU a0 is the attenuation measured in the aorta and HU Thr is the attenuation measured in the thrombus.
- Characteristics of these techniques (Why these techniques were used in the study?)
The study was designed to demonstrate a 5% difference in the sensitivity of endoleaks detection within the various data sets characterized by phantom size, tube voltage and the presence or lack of filtering. On the basis of the preliminary power analysis, one-way analysis of variance using two levels and 144 endoleaks in each data set can show the effect size of 5% with a power of 0.849 at a within-effect size of 0.2 and
The readers’ marks were compared with the location of the endoleaks in the phantom. Since the false-positive rate per image was very low (0.03 which is less than 0.05), analytical methods for data collected in a free-response manner could not be used. Therefore, analysis of variance for repeated measurements with post-hoc tests and Friedman analysis of variance were used to compare findings at various tube energies in the same phantom with or without a noise filter.
The effect of size and density of endoleaks, tube energy, phantom size, and noise filter on the number of true-positive findings, diagnostic confidence, subjective noise, and image quality was analyzed by using a general linear model. Interobserver agreement was assessed by analyzing the endoleaks detection in a lesion-to-lesion manner and calculating the weighted factor k. Agreement between the readers was graded as: < 0.20:poor, 0.21-0.40:fair, 0.41-0.60: moderate, 0.61-0.8-: good, and 0.81-1.00: very good.
- Clinical imaging applications and PACS implementation
The findings showed that the 3D ORA filter not only improved image quality at 80- and 100- kVp in all investigated patient size but also helped detect more lesions in simulated large patients at 100 kVp.
In simulated intermediate-sized patients, energy reduction from 120 to 100 kVp and from 100 to 80 kVp did not decrease image quality when images with reduced kilovoltage were filtered (P = 0.2692 and P > 0.99, respectively).
Readers detected more endoleaks on the filtered 100-kVp images than on the nonfiltered images in simulated large patients than on the nonfiltered images in simulated large patient (83 vs 75 lesions, P = 0.041). The number of detected endoleaks and the confidence rate were similar at 100 kVp with a filter and at 120 kVp in simulated large patients (P = 0.339 and P = 0.211, respectively).
Conclusion: In a phantom, the nonlinear noise filter can prevent decreased image quality with use of 80- and 100-kVp abdominal multidetector CT angiography at a wide range of simulated body weights and may facilitate a better detection rate of endoleaks in heavy patients.
Because of the 3D ORA does not degrade spatial resolution or require raw image data and can be used with DICOM images retrospectively in a relatively short time, it can be used in the clinical application. Additionally the authors suggest that the results may be applicable to the detection of small hyper vascular tumors in the liver or pancreases at low tube energy if the dependence of contrast enhancement on the total iodine load in abdominal solid organs is considered and if the CT protocol is adjusted accordingly.
