Nonlinear Gaussian Filter Research Articles

Feasibility of MDCT Colonography in Ultra-Low-Dose Technique in the Detection of Colorectal Lesions: Comparison with High-Resolution Video Colonoscopy

The purpose of this study was to assess the feasibility of MDCT colonography in an ultra-low-dose technique in the detection of endoluminal colonic lesions compared with high-resolution video colonoscopy. After standard bowel cleansing, 137 patients (77 men, 60 women; mean [+/- SD] age, 57.1 +/- 11.3 years) underwent high-resolution video colonoscopy within 2 hr after ultra-low-dose MDCT colonography had been performed. Ultra-low-dose MDCT colonography was performed with patients in the supine position only using 10 mAs (effective weighted CT dose index, 0.94 mGy). After mathematic noise reduction by nonlinear gaussian filter chains, using dedicated software (ECCET), images were analyzed by two blinded observers in simultaneously displayed interactive 2D and 3D modes. Findings of ultra-low-dose MDCT colonography were compared with the results obtained with high-resolution video colonoscopy. Calculated effective doses were 0.7 and 1.2 mSv for men and women, respectively. Ultra-low-dose MDCT colonography detected 84 (62%) of 135 lesions: 11 (78.6%) of 14 large polyps (> 10 mm), 12 (85.7%) of 14 medium polyps (9.9-5 mm), and 61 (57%) of 107 small polyps (< 5 mm). On a patient-by-patient basis, an overall sensitivity of 70.3% with a specificity of 80.8% was calculated. False-positive findings were seen mostly for small lesions (eight medium and 29 small lesions). Two of the three false-negative lesions were retrospectively detected in contrast-enhanced cleansing fluid; one was a flat lipoma not detectable on ultra-low-dose MDCT colonography. Despite an effective dose of approximately 1 mSv, MDCT colonography using an ultra-low-dose technique performs as well as MDCT colonography with a standard dose, according to published data. After mathematic noise reduction, 82% of polyps larger than 5 mm can be detected.

The combined effect of nonlinear filtration and window size on the accuracy of tissue displacement estimation using detected echo signals

In cardiac elastography, the regional strain and strain rate imaging is based on displacement estimation of tissue sections within the heart muscle carried out with various block-matching techniques (cross-correlation, sum of absolute differences, sum of squared differences, etc.). The accuracy of these techniques depends on a combination of ultrasonic imaging parameters such as ultrasonic frequency of interrogation, signal-to-noise ratio, size of a kernel used in a block-matching algorithm, type of data and speckle decorrelation. In this paper, we discuss the possibility to enhance the accuracy of the displacement estimation via nonlinear filtering of B-mode images before block-matching operation. The combined effect of a filter algorithm and a kernel size on the accuracy of the displacement estimation is analyzed using a 36-frame sequence of grayscale B-mode images of a human heart acquired by an ultrasound system operating at 1.77 MHz. It is shown that the nonlinear filtering of images enables to obtain the desired accuracy (less than one pixel) of the displacement estimation with smaller kernels than without filtering. These results are obtained for two filters––an adaptive anisotropic diffusion filter and a nonlinear Gaussian filter chain.

Noise reduction in PET attenuation correction using non-linear Gaussian filters

In a PET study, shortening of transmission scan time is highly desired for improving patient comfort and increasing scanner throughput. It necessitates a method that reduces statistical noise in attenuation correction factors (ACFs). The authors have evaluated non-linear Gaussian (NLG) filtering for smoothing transmission images reconstructed with filtered back-projection instead of using iterative reconstruction and segmentation methods. The NLG filtering operation is a variation of local weighted averaging in a neighborhood around a pixel, which weights are determined according to both distance in location and difference in pixel value. Several filtering steps with different NLG parameters can effectively reduce noise without losing structural information. The NLG smoothed transmission images are then forward projected to generate ACFs. Results with phantom and patient data suggested that the NLG filtering method is useful for attenuation correction using count-limited transmission data for both brain and whole-body PET studies.

Nonlinear Gaussian filtering approach for object segmentation

Gaussian filter kernels can be used to smooth textures for image segmentation. In so-called anisotropic diffusion techniques, the smoothing process is adapted according to the edge direction to preserve the edges. However, the segment borders obtained with this approach do not necessarily coincide with physical object contours, especially in the case of textured objects. A novel segmentation technique involving weighted Gaussian filtering is introduced. The extraction of true object masks is performed by smoothing edges due to texture and preserving true object borders. In this process, additional features such as disparity or motion are taken into account. The method presented has been successfully applied in the context of object segmentation to natural scenes and object-based disparity estimation for stereoscopic applications.

A nonlinear gaussian filter applied to images with discontinuities

We focus on a simple nonlinear filter with Gaussian weights for recovering the true underlying scene from images distorted by independent and identically distributed Gaussian noise. In some applications, like magnetic resonance (MR) brain scans, this model for the distortion is re-asonable. Apart from the good performance of this filter, the main advantage is that it is nearly instantaneous in processing time, which allows convenient interactive use. In this paper, we give a statistical justification of the filter by means of a weighted least squares interpretation and a Bayesian motivation. Furthermore, we describe how a smoothing parameter, which controls the degree of smoothness in the image, can be estimated from the observed image. Successful applications to both artificial and MR images are presented.