Elsevier

Ultrasonics

Volume 65, February 2016, Pages 177-193
Ultrasonics

Speckle filtering of medical ultrasonic images using wavelet and guided filter

https://doi.org/10.1016/j.ultras.2015.10.005Get rights and content

Highlights

  • A new de-speckling method based on the wavelet and Guided filter is proposed.

  • An improved threshold function based on the universal threshold function is propose.

  • A new shrinkage algorithm is designed.

  • The experiments explain the applicability of the proposed method for the medical ultrasound images.

Abstract

Speckle noise is an inherent yet ineffectual residual artifact in medical ultrasound images, which significantly degrades quality and restricts accuracy in automatic diagnostic techniques. Speckle reduction is therefore an important step prior to the analysis and processing of the ultrasound images. A new de-noising method based on an improved wavelet filter and guided filter is proposed in this paper. According to the characteristics of medical ultrasound images in the wavelet domain, an improved threshold function based on the universal wavelet threshold function is developed. The wavelet coefficients of speckle noise and noise-free signal are modeled as Rayleigh distribution and generalized Gaussian distribution respectively. The Bayesian maximum a posteriori estimation is applied to obtain a new wavelet shrinkage algorithm. The coefficients of the low frequency sub-band in the wavelet domain are filtered by guided filter. The filtered image is then obtained by using the inverse wavelet transformation. Experiments with the comparison of the other seven de-speckling filters are conducted. The results show that the proposed method not only has a strong de-speckling ability, but also keeps the image details, such as the edge of a lesion.

Introduction

Ultrasonic imaging is a powerful technique for viewing the internal anatomy (e.g., abdomen, breast, liver, kidney, and musculoskeletal). It is non-invasive, non-radioactive, convenient and efficient compared with other medical diagnosis methods. Thus, the clinical application of ultrasonic imaging technology has become more important, especially in observing the growth status of the fetus in pregnant women and the diagnosis of lesions of the abdominal organs.

However, like all coherent imaging methods, ultrasonic imaging has a main disadvantage, i.e., it is contaminated by speckle noise. Speckle noise is generated by interaction of the reflected waves from various independent scatters within a cell resolution [1]. The existence of speckle noise has significantly degraded the quality of ultrasound images and restricted the development of automatic diagnostic techniques. For clinicians, speckle noise has influenced their accurate diagnosis, especially for not very experienced doctors. Therefore, from the perspective of clinical application, de-speckling becomes an important step prior to the analysis and processing of ultrasound images, and many researchers are inspired to devote their efforts to this issue. It provides a technical incentive for the doctors to obtain more accurate diagnosis, and reduces the risk of misdiagnose.

A number of de-noising methods for medical ultrasound images have been proposed within scholarly literature. These can be simply classified into five categories, i.e. local adaptive filter, anisotropic diffusion filter, multi-scale filter, nonlocal means filter and hybrid filter [2]:

  • (1)

    The commonly used adaptive filters, e.g. the median [3], Bilateral [4] and SRBF filters [5] assume that speckle noise is essentially a multiplicative noise. Although the local adaptive filters have low algorithm complexity, they tend to blur images and cannot provide satisfactory de-noised results.

  • (2)

    The anisotropic diffusion filters include DTD [6], SUSAN_AD [7] and OSRAD [8]. Most of them have strong speckle noise suppression ability, but the de-noised results may create an over-smooth phenomenon.

  • (3)

    Nonlocal means filters are novel de-noising algorithms, such as OBNLM [2], PPB [9] and Guo et al. [10]. The Nonlocal means filters have better speckle noise removing effect, but the algorithm complexity of these filter are usually very high, and thus they cannot meet the real-time requirement of medical ultrasound imaging system.

  • (4)

    The multi-scale and wavelet based approaches have been widely used in signal de-noising due to the advantages of time–frequency analysis and multi-scale analysis, like the wavelet soft/hard threshold filter [11] and Andria filter [12].

  • (5)

    The aforementioned approaches can also be used in various combinations with each other in order to take advantage of the different paradigms each method provides. For example, the SAR-BM3-D’s filter [13] is the combination of the multi-scale filter with the nonlocal means filter.

Since the above methods do not achieve optimal balance between speckle suppression and feature preservation, it is desirable to develop a filter, which improves the performances of strong de-noising ability and better edge-preservation, as well as non-iterative and low algorithm complexity. A new de-noising method based on wavelet de-noising and guided filter is proposed in this paper, where the low algorithm complexity of the wavelet de-noising method and the strong de-noising ability of the guided filter are combined. Wavelet threshold de-noising is a classical wavelet de-noising method. It has higher efficiency and performs better than others in the processing of additive noise, which can satisfy general product demand. Since speckle noise in medical ultrasound images appears larger and “granular”, it is found in the experiments that speckle noise still exists in the low frequency sub-band of the wavelet domain. Therefore, the guided filter is implemented to filter speckle noise in the low frequency sub-band. Some authors had suggested the use of a bilateral filter [14]. The bilateral filter can maintain the edge information and has desirable de-noising performance, but it suffers from two main disadvantages: i.e., high complexity and “gradient distortion”, which cannot meet the real-time requirement of medical ultrasonic imaging systems. So He et al. [15] put forward the concept of the guided filter in 2010, which has greatly improved the de-noising performance and efficiency.

Therefore, advantages of the wavelet de-noising method and the guided filter will be combined in this paper. The main work of our proposed method is as follows: On the one hand, in order to balance between speckle suppression and feature preservation, an improved wavelet threshold function related to the layer number of wavelet decomposition is designed. In addition, according to the statistical models of noise and signal in the wavelet domain, where the wavelet coefficients of noise-free signal and speckle noise are modeled as generalized Gaussian distribution and Rayleigh distribution respectively, Bayesian maximum a posteriori estimation is applied to develop a new shrinkage algorithm. On the other hand, in order to filter the larger speckle noise in the low frequency sub-band, the guided filter is implemented. In the experiments, the proposed method is compared with other seven de-speckling filters, and the results show that the proposed method not only has a strong de-speckling ability, but also maintains the image details, such as the edge of the lesion, which can enhance the quality of medical ultrasound images.

The rest of the paper is organized as follows: The model of speckle noise in medical ultrasound images is introduced in Section 2. Section 3 demonstrates the approach of the proposed de-noising method. In Section 4, simulation and experiment studies are presented. Conclusions are summarized in Section 5.

Section snippets

The model of speckle noise in medical ultrasound image

It is very important to understand the statistical model of speckle noise before the noise removal in medical ultrasonic images can be considered. The ultrasonic signal is emitted into the human tissue and the reflected ultrasonic signal (envelope signal) is obtained and processed by the receiving circuit of ultrasonic imaging equipment, where the main envelope signal is the reflected signal of the human body tissues and organs. The final ultrasonic envelope signal obtained consists of two

Speckle reduction of ultrasound images by wavelet and guided filter

According to the requirements of de-noising for medical ultrasound images, a new de-speckling method based on the wavelet shrinkage algorithm with guided filter is proposed in this section. The three innovation points are as follows:

  • (1)

    In order to balance between speckle suppression and feature preservation, an improved wavelet threshold function related to the layer number of wavelet decomposition is designed based on the universal threshold value function.

  • (2)

    According to the statistical models of

Experimental studies of the proposed method

From the theoretical description of the above sections, an improved de-noising method is proposed in this paper based on the characteristics of the speckle noise in medical ultrasonic images. This method has superiority in theory compared with the basic de-noising methods. So in order to further verify this method, more comparison experiments will be carried out. In this paper, the experiments are divided into two branches: The speckle noise simulation experiment and the medical ultrasound

Conclusions

Aiming at the problem of speckle noise in medical ultrasound images, an improved de-speckling method is proposed in this paper, which is based on the wavelet de-noising method and guided filter. Three contributions in this paper are made:

  • (1)

    In order to balance between speckle suppression and feature preservation, a new wavelet threshold function related to the layer number of wavelet decomposition is designed based on the universal threshold function. The new threshold function not only weakens

Acknowledgments

The authors would like to thank the anonymous reviewers for their valuable comments and suggestions that will help improve the manuscript. The authors would also like to thank the people who provide the MATLAB code or executable file. The work is partially supported by National Natural Science Foundation of China (60974042).

References (26)

  • C. Deledalle et al.

    Iterative weighted maximum likelihood denoising with probabilistic patch based weights

    IEEE Trans. Image Proces.

    (2009)
  • D.L. Donoho et al.

    Ideal spatial adaptation via wavelet shrinkage

    Biometrika

    (1994)
  • S. Parrilli et al.

    A nonlocal SAR image denoising algorithm based on LLMMSE wavelet shrinkage

    IEEE Trans. Geos. Remo. Sens.

    (2012)
  • Cited by (58)

    • Ultrasonic signal enhancement for coarse grain materials by machine learning analysis

      2021, Ultrasonics
      Citation Excerpt :

      This enables identification between the desired and irrelevant information for transient signals, even overlapping each other in frequency. The WPT has attracted considerable attention for decorrelating noise in the fields of signal processing, image processing, speech and audio processing, and geophysics [2–7]. A denoising approach using two-dimensional stationary wavelet transform with analytic wavelet thresholding was proposed for suppressing random and wedge noises in the ultrasonic angle-beam pulse-echo setup by Mohammad R. Hoseini et al. [8].

    • A review on medical image denoising algorithms

      2020, Biomedical Signal Processing and Control
      Citation Excerpt :

      The spatial domain filters, in general, tend to blur the images even though they seem to have low computational complexity and hence do not cater to the needs of a medical practitioner. Filtering based methods, therefore tend to compromise between speckle suppression and image preservation [88,89]. Low-rank techniques are another area to be utilised for the denoising process.

    • Analysis of quantum noise-reducing filters on chest X-ray images: A review

      2020, Measurement: Journal of the International Measurement Confederation
      Citation Excerpt :

      Furthermore, it overcomes the gradient reversal artifact that appears in the case of the bilateral filter [31]. It has been widely applied in the field of image de-noising [32], interactive image segmentation [33], HDR compression, de-hazing, image matting, etc. In this study, the input image itself is used as a guidance image.

    View all citing articles on Scopus
    View full text