Noise-reduction approaches to single-shot dual-energy imaging with a multilayer detector

Abstract

The single-shot dual-energy (DE) method using a sandwich-like multilayer detector generally suffers from noise in the resultant DE images due to the poor quantum noise in images obtained from the rear detector layer and its amplification due to the subtraction operation in DE reconstruction. The use of interlayer metal filter can further increase noise in DE images by increasing the rear-detector quantum noise. We have adopted several noise-reduction algorithms to the singleshot DE reconstruction, which included the Gaussian-filtering noise reduction (GNR), the medianfiltering noise reduction (MNR), and the anti-correlated noise reduction (ANR). We assessed the effectiveness of noise-reduction methods by investigating noise and dose-normalized contrast-tonoise ratio for a mouse-mimicking phantom consisting of aluminum bone and polyurethane soft tissues. Noise-power spectrum was also measured to investigate correlation noise. Overall, the ANR showed the best performance. At some extreme imaging technique conditions, such as a lower tube voltages and a larger spectral energy separation using thick copper sheets between the front and rear detector layers, the MNR outperformed the ANR. The performance of GNR was nearly similar to that of MNR. The results were well reflected into demonstration bone images obtained for a postmortem mouse. Although the ANR was effective to reduce noise in the single-shot DE imaging using the sandwich detector, multivariate optimization of ANR parameters with respect to imaging tasks and imaging techniques is remained as a future study. K: Medical-image reconstruction methods and algorithms, computer-aided diagnosis; X-ray radiography and digital radiography (DR)

Soft-tissue radiography using multi-scale convolutional neural networks

“…As shown in column (1) of figure 4, we observe edge artifacts in images obtained from the weighted-subtraction operation. In general, those artifacts result from when subtracting two images with different spatial-resolution characteristics such as the single-shot DEI using a sandwich detector, which consists of two different detector layers [17,18], and the DEI employing noise reduction [19]. Noise reduction in DEI using the anti-correlation property between two energy images [20] is typically known to have apparent edge artifacts, depending on the kernel size [21].…”

Study on Dual-Energy Signal and Noise of Double-Exposure X-Ray Imaging for High Conspicuity

“…A Gaussian filter was applied with a Gaussian distribution to a HE image, which is often used to reduce the noise generated by a normal, or probability distribution. The GNR is expressed as follows [7]: (8) where fG defined as, (9) which implies the kernel size of the filter and σG denotes the standard deviation of the Gaussian function. The kernel size for the GNR was 7 × 7.…”

“…To reduce noise from a HE image, the median filter of nonlinear filtering was used, and an MNR with a 7 × 7 kernel size is expressed as follows [9]: (10) 3) Anti-correlated filter…”

Development of Artificial Intelligence-Based Dual-Energy Subtraction for Chest Radiography

“…In the one-shot method, two images are simultaneously obtained at different energies by placing a thin copper plate between two imaging detectors [20]. Because of the copper plate, noise characteristics can deteriorate the quality of high-energy images [21]. In the two-shot method, on the other hand, X-ray exposure is carried out twice at different energies [22].…”