Frequency decomposition and compounding of ultrasound medical images with wavelet packets

Abstract

Ultrasound beams propagating in biological tissues undergo distortions due to local inhomogeneities of the acoustic parameters and the nonlinearity of the medium. The spectral analysis of the radio-frequency (RF) backscattered signals may yield important clinical information in the field of tissue characterization, as well as enhancing the detectability of tissue parenchymal diseases. In this paper, we propose a new tissue spectral imaging technique based on the wavelet packets (WP) decomposition. In a conventional ultrasound imaging system, the received echo-signals are generally decimated to generate a medical image, with a loss of information. With the proposed approach, all the RF data are processed to generate a set of frequency subband images. The ultrasound echo signals are simultaneously frequency decomposed and decimated, by using two quadrature mirror filters, followed by a dyadic subsampling. In addition, to enhance the lesion detectability and the image quality, we apply a nonlinear filter to reduce noise in each subband image. The proposed method requires simple additional signal processing and it can be implemented on any real-time imaging system. The frequency subband images, which are available simultaneously, can be either used in a multispectral display or summed up together to reduce speckle noise. To localize the different frequency response in the tissues, we propose a multifrequency display method where three different subband images, chosen among those available, are encoded as red, green, and blue intensities (RGB) to create a false-colored RGB image. According to the clinical application, different choices can evidence different spectral proprieties in the biological tissue under investigation. To enhance the lesion contrast in a grey-level image, one of the possible methods is the summation of the images obtained from narrow frequency subbands, according to the frequency compounding technique. We show that by adding the denoised subband images created with the WP decomposition, the contrast-to-noise ratio in two phantom images is largely increased.

Hybrid Filter Based on Neural Networks for Removing Quantum Noise in Low-Dose Medical X-ray CT Images

“…The CNR is defined as where µ u and σ u are the mean and the standard deviation, respectively, computed in an undesired region of interest (UROI), such as background. Both MSR and CNR measurements are proportional to the medical image quality [20]. Table 2 shows the MSR and CNR values of the proposed filter and others.…”

An experimental study of ultrasonic imaging with a reduced number of array elements using the envelope method

“…The parameters are the same as in [21], where four-level WP decomposition is assumed using a Daubechies db3 wavelet to decompose each signal into sixteen sub-band signals. Next, a soft threshold algorithm is applied to these wavelet components, where the threshold t h ¼ s e ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 log 2 N p is determined with a data length N ¼ 512 and a risk-related parameter s e ¼ 0:1.…”

“…In medical imaging, both the media and targets are inhomogeneous, and they generate speckle noise. To address this problem, the wavelet-based frequency decomposition imaging method was proposed by Cincotti et al [21]. This method employs wavelet packets (WP) to decompose signals into multiple sub-band signals, and is one of the frequency compounding imaging methods.…”

Ultrasound Image Denoising by Spatially Varying Frequency Compounding

“…We chose blocks in which there is no meaningful object detail. 4 Covariance depends on both lateral and radial lags. Furthermore, radial and lateral correlations differ.…”

“…Most past approaches for denoising ultrasound images have used standard image reconstruction tools, such as weighted median filter [9], wavelet based methods [4] [5], Gaussian non-linear filters [3] and anisotropic diffusion [14]. All these methods essentially blur the image.…”