No abstract

In recent decades, X-ray CT imaging has become more important as a result of its high-resolution performance. However, it is well known that the X-ray dose is insufficient in the techniques that use low-dose imaging in health screening or thin-slice imaging in work-up. Therefore, the degradation of CT images caused by the streak artifact frequently becomes problematic. In this study, we applied a Wiener filter (WF) using the universal Gaussian mixture distribution model (UNI-GMM) as a statistical model to remove streak artifact. In designing the WF, it is necessary to estimate the statistical model and the precise co-variances of the original image. In the proposed method, we obtained a variety of chest X-ray CT images using a phantom simulating a chest organ, and we estimated the statistical information using the images for training. The results of simulation showed that it is possible to fit the UNI-GMM to the chest X-ray CT images and reduce the specific noise.

No abstract

Purpose: Although the signal-to-noise ratio (SNR) currently used in the field of medical X-ray CT is utilized for local image evaluation in a linear system, it is not used as a comprehensive evaluation index for an entire image. Additionally, since X-ray CT cannot produce a noiseless image for obtaining the signal power required to calculate the SNR, it is impossible to calculate SNR precisely even applying the conventional method. To resolve these problems, we propose SNR * , which is a new method for calculating the estimated value of SNR that can evaluate an entire image even when the original image cannot be obtained. Methods: First, we obtained SNR * using the signal power and noise power calculated respectively from covariance and the difference in the pair of observed images, which are noise-containing images scanned under the same imaging conditions. Next, we verified the error and the accuracy of SNR * . Third, we demonstrated the behavior and accuracy of the SNR * applied to the actually observed image. Results: In the verification experiment, the relative error of SNR * concerning the true value was 0.06% or less, and the coefficient of variation value of SNR * in the demonstration experiment was 0.015 or less, which denoted the accuracy of SNR * . Conclusion: The proposed method realizes SNR measurement even in cases in which only observed images can be obtained and original images cannot be obtained, such as Xray CT images.

No abstract