Elsevier

Medical Image Analysis

Volume 16, Issue 6, August 2012, Pages 1101-1112
Medical Image Analysis

Ultrasound confidence maps using random walks

https://doi.org/10.1016/j.media.2012.07.005Get rights and content

Abstract

Advances in ultrasound system development have led to a substantial improvement of image quality and to an increased use of ultrasound in clinical practice. Nevertheless, ultrasound attenuation and shadowing artifacts cannot be entirely avoided and continue to challenge medical image computing algorithms. We introduce a method for estimating a per-pixel confidence in the information depicted by ultrasound images, referred to as an ultrasound confidence map, which emphasizes uncertainty in attenuated and/or shadow regions. Our main novelty is the modeling of the confidence estimation problem within a random walks framework by taking into account ultrasound specific constraints. The solution to the random walks equilibrium problem is global and takes the entire image content into account. As a result, our method is applicable to a variety of ultrasound image acquisition setups. We demonstrate the applicability of our confidence maps for ultrasound shadow detection, 3D freehand ultrasound reconstruction, and multi-modal image registration.

Highlights

► We estimate a per-pixel confidence for ultrasound images. ► Confidence is estimated with a random walks framework. ► Shadow regions are reliably detected. ► The contrast is improved for ultrasound reconstruction. ► The accuracy and capture range is improved for CT–US registration.

Introduction

Ultrasound imaging is an essential part of clinical routine offering real-time imaging of patient anatomy. In the last decade advances in ultrasound transducer and system development have led to a substantial improvement of ultrasound image quality and to an ever increasing use in clinical practice. However, ultrasound attenuation and shadow artifacts are still unavoidable, and continue to challenge not only physicians but also engineers in the field of medical image computing. Algorithms for ultrasound image processing are confronted with reduced image contrast and loss of anatomical structures in attenuated and/or shadowed image regions (Leroy et al., 2004). To overcome these challenges they commonly apply ultrasound specific-constraints and prior knowledge to their underlying methods (Noble and Boukerroui, 2006). Moreover, the accuracy and robustness of ultrasound image processing methods commonly depends on their ability to cope with attenuated and/or shadowed regions. Consequently, approaches that provide knowledge about such regions to other algorithms are of high relevance for different applications, including: ultrasound segmentation, registration, and reconstruction.

In this paper we introduce a novel method for estimating the uncertainty in ultrasound images caused by attenuation and/or shadowing. More specifically, we introduce a per-pixel confidence, i.e., a confidence map, calculated from the ultrasound image content. Our main novelty lies in the confidence estimation process, which is performed within a random walks framework; previously introduced for image segmentation by Grady (2006). The desired confidence map is obtained by answering the following question: What is the probability that a random walk starting from a pixel would be able to reach the virtual transducer elements, given the ultrasound image and ultrasound specific constraints? For this we assume that the likelihood of ultrasound transmission is directly related to the confidence of the image information, i.e., the lower the ultrasound transmission the less the confidence we have in the image information. The ultrasound specific constraints are derived from domain specific knowledge for ultrasound and include: transmission modeling, depth dependent attenuation and ultrasound scanline constraints. A first introduction of this method was presented in a shortened form for analyzing the confidence in Intravascular Ultrasound Radio-Frequency data in correlation with co-registered histology (Karamalis et al., 2012).

We will further discuss the benefits of a confidence measure in the related work (Section 2), together with other approaches that have used measures of confidence, probability, information content, and image quality to improve the outcome of their target application. Subsequently, details of our method will be provided in Section 3, including the necessary adjustments to the random walks framework, the modeling assumptions, and implementation details. In Section 4 we analyze the free parameters that are involved in the confidence estimation and evaluate the computed estimate for regions of a priori known confidence. Our confidence maps are not intended to be used independently, but rather to be integrated into other ultrasound image processing methods; providing knowledge of possible attenuated and/or shadowed regions. Consequently, we demonstrate the applicability of our approach for common applications, including: shadow detection (Section 5.1), 3D freehand ultrasound reconstruction (Section 5.2), and multi-modal CT (Computer Tomography) – US (Ultrasound) image registration (Section 5.3). We discuss the results in Section 6 and conclude by summarizing and pointing out future works in Section 7.

Section snippets

Related work

The idea/concept of an ultrasound confidence map has been addressed in the literature with different approaches and in different contexts. Methods that estimate ultrasound attenuation can be used to detect shadowed and/or attenuated areas, thus, providing knowledge about unreliable image regions. The estimation of ultrasound transmission allows algorithms to mask out unreliable regions, like shadows, in order to improve the accuracy and robustness of their underlying methods. More generally,

Methods

Random walks for image segmentation, introduced by Grady (2006), has become a widely used approach in the computer vision and the medical image segmentation community. Subsequently, the algorithm was applied to several other computer vision and graphics problems including alpha-matting (Grady et al., 2005a), mesh segmentation (Zhang et al., 2010), mesh denoising (Sun et al., 2008) and stereo matching (Shen et al., 2008).

In this work we propose the use of the random walk framework for ultrasound

Evaluation

We provide qualitative results in form of confidence maps obtained from ultrasound images for different transducer geometries and anatomies, as shown in Fig. 2. Before we move onto the other results we would first like to discuss the free parameters of our method.

Applications

We demonstrate the applicability of the confidence maps for different applications including: shadow detection, 3D freehand ultrasound reconstruction, and CT–US registration.

Discussion

The introduced confidence maps are not intended to be used independently, but rather in combination with other ultrasound image computing/analysis algorithms; providing an estimate of uncertainty in attenuated and/or shadowed regions. Therefore, we demonstrated the benefit of integrating confidence maps into ultrasound shadow detection, 3D freehand ultrasound reconstruction, and US-CT registration. Our goal was not to demonstrate superior segmentation, registration, or reconstruction results

Conclusion

We presented a generic and novel method for estimating a per-pixel confidence in ultrasound images, which we denoted ultrasound confidence map, that emphasizes the uncertainty in attenuated and/or shadowed regions. The main novelty of our approach is the formulation of the confidence estimation problem within a random walks framework. Domain specific knowledge of ultrasound was integrated into the framework, including: the relation of possible ultrasound transmission to image intensities, depth

Acknowledgments

The authors thank Nicolas Brieu and Maximilian Baust for valuable comments, Leo Grady for valuable comments and his help with the revision of this article, Seyed-Ahmad Ahmadi for his help with the ultrasound calibration, and José Gardiazabal for his help with the hardware setup. Further thanks to the anonymous reviewers, whose constructive feedback helped in extending and improving the original manuscript. This work was partly funded by the European projects: “PASSPORT” – Grant Agreement no:

References (41)

  • X. Sun et al.

    Random walks for feature-preserving mesh denoising

    Computer Aided Geometric Design

    (2008)
  • W. Wein et al.

    Automatic CT–ultrasound registration for diagnostic imaging and image-guided intervention

    Medical Image Analysis

    (2008)
  • D. Barratt et al.

    Self-calibrating 3D-ultrasound-based bone registration for minimally invasive orthopedic surgery

    IEEE Transactions on Medical Imaging

    (2006)
  • V. Behar et al.

    Statistical analysis of image quality in multi-angle compound imaging

  • N. Bell et al.

    Implementing sparse matrix-vector multiplication on throughput-oriented processors

  • R.S.C. Cobbold

    Foundations of Biomedical Ultrasound

    (2007)
  • L. Grady

    Random walks for image segmentation

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (2006)
  • L. Grady

    A lattice-preserving multigrid method for solving the inhomogeneous poisson equations used in image analysis

  • L. Grady et al.

    Weights and topology: a study of the effects of graph construction on 3D image segmentation

  • Grady, L., Schiwietz, T., Aharon, S., Westermann, R., 2005a. Random walks for interactive alpha-matting. In:...
  • Cited by (0)

    View full text