Original contribution
Ultrasound Compounding With Automatic Attenuation Compensation Using Paired Angle Scans

https://doi.org/10.1016/j.ultrasmedbio.2006.09.012Get rights and content

Abstract

Variations in attenuation in tissue can result in shadowing and enhancement in ultrasound images. Angular compounding of ultrasound images by lateral beam-steering can be used to improve delineation of structures, but causes such shadows and enhancements to appear in less recognisable forms. We present an algorithm which uses lateral beam-steering to produce compounded images with significantly reduced artefacts, by considering the response from equal and opposite angles. This is compared to several other alternative algorithms for attenuation estimation, some of them embedded for the first time in a multi-angle framework. Algorithms are tested on simulated and in vitro data in 2D and 3D contexts. Gain variations across all observed shadows and enhancements are reduced to below 5 dB. The new algorithm is as good as the best alternative on all data sets tested, and is straightforward to implement. We end by discussing further work required to relax the necessary assumptions in order to achieve a similar level of performance on in vivo data.

Introduction

Medical ultrasound images contain many artefacts due to the complex nature of sound transmission and reflection in anatomical structures. Shadowing and enhancement, for instance, are the result of variations in attenuation throughout the image. This can be compensated in the axial direction by a set of time-gain sliders, which control the gain in lateral bands across the image. Gain variations in the lateral direction remain, however, to appear as overcompensated bright patches after regions of low attenuation (enhancement) or undercompensated dark patches after regions of high attenuation (shadowing).

Since shadows and enhancements are in effect indicators of relative attenuation in overlying regions, something which is otherwise not displayed, they can in some cases have clinical significance. For instance, this has been demonstrated in detecting liver disease (Bevan and Sherar, 2001) or certain tumours (Tu et al., 2003) or even for monitoring temperature change (Tyréus and Diederich, 2004). However, in other cases they can simply be confusing. This is particularly the case for 3D data, where the visualisation planes are not in general along the direction of insonification. In such planes, shadows or enhancements can appear without the corresponding anatomy that generated them.

In any case, it would seem logical to display attenuation effects separately from signal backscatter (or reflection), which is the main component of ultrasound images. Certainly this would ease the interpretation of ultrasound images and make downstream processing (for instance segmentation) more reliable. The estimated attenuation can then be displayed as a separate image rather than having to infer it from artefacts in an image essentially of ultrasound backscatter.

One technique for reducing shadowing and enhancement is to attempt to estimate the attenuation independently from the backscatter. Having estimated the attenuation at all points in an image, it is straightforward to adjust the image for this known attenuation, therefore removing artefacts. Such approaches have been reviewed in a previous paper (Treece et al., 2005). All rely on certain assumptions regarding the ultrasound signals. The algorithm of Hughes and Duck (1997) deduces attenuation directly from the backscattered signal by assuming that the attenuation is directly proportional to the backscatter. Flax et al. (1983) assume that the ultrasound pulse has a broadly Gaussian spectrum, and use the shift in centre frequency to estimate attenuation. Knipp et al. (1997) compare the backscattered spectrum with a calibration spectrum to estimate attenuation. Both these last methods require the scattering spectrum of the sample to be the same as in a calibration object.

Angular compounding, where the ultrasound beam is steered laterally to various angles and the resulting images averaged, was first proposed over twenty years ago (Berson et al., 1981), but has recently become more generally accessible (Entrekin et al 1999, Entrekin et al 2001, Jesperson et al 1998). The main benefit of such compounding is to increase the signal to speckle ratio and reduce the dependency of reflection from planar interfaces on relative angle to the transducer. However, since shadowing and enhancements will always lie in the direction of insonification, angular compounding also has the subsidiary effect that these artefacts are blurred by the compounding procedure. Looking ahead to Fig. 5a, b reveals some in vitro examples—the strength of the shadow is not substantially reduced by compounding, but the appearance is somewhat different than in a conventional B-scan.

Several authors have recently attempted to use angular compounding on quantities other than signal magnitude, for instance attenuation (Tu et al., 2005), strain (Rao et al., 2006) and temperature (Pernot et al., 2004) estimates. In all cases, ultrasound scans at each insonification angle have simply been regarded as relatively independent estimates, which can be averaged to increase the information content. Wilhjelm et al. (2004) recently investigated a slight variation whereby various forms of median and maximum filters were used instead of a simple average from angular scans.

To our knowledge, the only investigation that aimed to deduce separable attenuation and backscatter coefficients from angled data was conducted nearly thirty years ago by Duck and Hill (1979). In this study, objects were scanned across a wide range of angles (typically 180° or more), and the attenuation and backscatter maps reconstructed by a forward iterative procedure. The wide range of angles was achieved by moving a conventional probe around objects placed in a water bath. We follow the approach of Duck and Hill (1979) in this paper in that insonification from a variety of known angles is used to deduce the attenuation, rather than simply averaging results from each angle. However, the range of angles is much more restrictive, in order to allow the approach to be used in steered angular compounding, and we present a direct rather than iterative solution to the problem. Nevertheless, the assumptions we make (and the resulting restrictions on the method) are similar in both cases.

It will be shown that, under certain assumptions, it is possible to calculate lateral variations in attenuation in a sample from the signal envelope of a pair of scans from equal and opposite steered angles. This information can be used to provide a compounded backscatter image free from shadows and enhancements. For comparison, we also embed the various competing attenuation estimation approaches in an angular compounding framework, in a similar manner to Tu et al. (2005).

We start in the following section by outlining a simple model for attenuation and backscatter, which allows us to develop the theory required for attenuation from paired-angle scans. We then outline various practical considerations necessary for implementing this theoretical result. In the following section, the new algorithm is compared with other forms of attenuation correction, with and without compounding, using simulated and in vitro data. Finally, results of these experiments are discussed and conclusions are drawn. We also discuss some future research avenues that may provide greater applicability to in vivo data.

Section snippets

Model for attenuation and backscatter

We deliberately adopt a simple model for the attenuation and backscatter in tissue. Clearly, real in vivo data will depart from such a model in a variety of ways. Nevertheless, a simple model allows us to develop the method, which we then test on various forms of simulated and in vitro data. We discuss in the Conclusions section how this might be extended to increase the applicability to in vivo data.

We assume that, at least over a small range of insonification angles, both the backscatter and

Experiments and Results

The algorithm was tested on simulation data and two different ultrasound phantoms. In all cases, compounding was performed by steering the ultrasound beam to 15 angles varying from −14° to 14° in steps of 2°. Ultrasound visualisations were generated by one of five methods:

  • BSCAN. The conventional B-scan display of received amplitude.

  • SCATTER. Attenuation correction using the method of Hughes and Duck (1997) in which the attenuation is assumed to be everywhere proportional to the backscatter.

Discussion

It is clear from these results that, although most of the algorithms perform well on some data, PAIRED is the most consistent. Although there are many assumptions behind this algorithm, they are in fact less strong than those behind the alternatives. For instance, PAIRED does require isotropic scattering, however, SPEC_DIV and CENTRE_FREQ require scattering of exactly the same type as in the calibration phantom, i.e., both isotropic and with the same frequency distribution. In this regard,

Conclusions

We have shown that it is possible, at least on simulated and in vitro data, to estimate variations in lateral attenuation using scans from equal and opposite steered angles. These can be used to generate spatially compounded images with significantly reduced shadowing and enhancement artefacts. The algorithm performs as well as the best alternatives and is more consistent over a range of data. Furthermore, it is relatively easy to implement and executes in real time.

There are a variety of

Acknowledgments

GT is supported by an EPSRC/RAEng Postdoctoral Fellowship. Dynamic Imaging Ltd. provided a modified ultrasound machine with low-level access to the control software. This enabled the acquisition of analogue RF and digital amplitude ultrasound data in real time and adjustments to the focusing required for beam steering.

References (22)

  • J.A. Jensen

    Field: A program for simulating ultrasound systems

    Med Biol Engin Comput

    (1996)
  • Cited by (12)

    • Mechanical characterization of abdominal aortas using multi-perspective ultrasound imaging

      2021, Journal of the Mechanical Behavior of Biomedical Materials
      Citation Excerpt :

      Several methods were introduced to overcome these limitations. Most methods aim to improve image quality and motion tracking (Foster et al., 1993; Righetti et al., 2003), whereas other methods use beam steering (Hansen et al, 2009, 2010; Treece et al., 2007). However, these beam steering techniques are inadequate for vessels situated deeper in the body (e.g. the abdominal aorta), due to small overlap between the steered ultrasound beams.

    • Enhancing Lateral Contrast Using Multi-perspective Ultrasound Imaging of Abdominal Aortas

      2021, Ultrasound in Medicine and Biology
      Citation Excerpt :

      Most commonly, these methods are applied to improve image quality as well as ultrasound motion tracking, where accurate measurement of the 2-D deformation field is hampered by the low lateral resolution (Foster et al. 1993; Righetti et al. 2003). Methods have been proposed where the ultrasound beam is steered under multiple angles (Shankar 1986; Treece et al. 2007; Hansen et al. 2009). With the steering of the ultrasound beam, the angle of ultrasound travel direction changes.

    • Characterization of the normal pulmonary surface and pneumonectomy space by reflected ultrasound

      2011, Journal of Ultrasound
      Citation Excerpt :

      This trend could further modify normal visualization of anatomic tissues, eliminating the interference and misleading effects of the numerous artifacts that are normally present in the thoracic cavity [22]. Artifacts are “imaging errors” caused by physical processes that affect the ultrasound beam and in some way alter the basic assumptions the operator makes about the beam [23]. For the time being, pleuropulmonary ultrasonography can be considered a complementary method for diagnosing several chest diseases; it is also an optimal guide during percutaneous procedures involving the aspiration of pleural effusions or the biopsy of peripheral pleural and lung consolidations [24,25].

    • Investigation into the fusion of multiple 4-D fetal echocardiography images to improve image quality

      2010, Ultrasound in Medicine and Biology
      Citation Excerpt :

      These techniques can be divided into those that seek to improve a single acquisition, such as the use of second harmonics, frequency blending or speckle reduction methods and those that seek to combine information from multiple acquisitions, such as image persistence or cross-beam approaches. Combining multiple images has been used to improve the information available in a number of contexts, including: 2-D ultrasound image enhancement using temporal persistence (Evans and Nixon 1996) or spatial compounding (Whatmough et al. 2006; Treece et al. 2007); 3-D image creation using freehand scanning (Xiao et al. 2002; Rohling et al. 1996); and fusion of ultrasound with other modalities for interoperative guidance (Porter et al. 2001; Wein et al. 2005). In all these contexts, multiple images containing complementary information are used either to enhance the information available to the clinician, or to create a single image of improved quality.

    • Ultrasound Image Optimization (Knobology): B-Mode

      2020, Ultrasound International Open
    View all citing articles on Scopus
    View full text