Original ContributionContrast-Ultrasound Dispersion Imaging for Prostate Cancer Localization by Improved Spatiotemporal Similarity Analysis
Introduction
Prostate cancer is the most common form of cancer in men in the United States, representing 29% and 9% of all cancer diagnoses and deaths, respectively (American Cancer Society 2012). Treatment often involves a radical prostatectomy, which carries the risk of severe permanent side effects like incontinence and impotence (Bangma et al. 2007). This risk could be reduced by focal therapies (Polascik and Mouraviev 2008), but their use is complicated by diagnostic limitations. In fact, diagnosis requires systematic biopsies, in which the prostate is uniformly sampled up to more than 16 times by a core needle. Imaging methods could significantly improve the current situation by enabling better patient stratification, biopsy targeting and focal therapy guidance, but are not yet available.
Angiogenesis is a key prognostic indicator for prostate cancer imaging, especially because of its correlation with cancer aggressiveness and the risk of developing metastasis (Brawer 1996; Weidner et al. 1993). This biochemical process leads to the formation of a dense microvascular network supporting the growth of prostate cancer beyond 1 mm3 (Brawer 1996). Differences in the microvascular architecture are characterized by an increased microvascular density as well as a higher tortuosity and permeability of the vessel wall (Bigler et al. 1993).
Detection of angiogenesis by assessment of tissue perfusion using techniques such as dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), Doppler ultrasound imaging and dynamic contrast-enhanced ultrasound (DCE-US) imaging has been proposed (Russo et al. 2012; Seitz et al. 2009; Smeenge et al. 2011). Given its use in biopsy guidance and its real-time availability at the bedside, ultrasound offers a practical and cost-effective alternative to magnetic resonance imaging for prostate imaging.
Dynamic contrast-enhanced ultrasound is especially interesting because of its ability to obtain flow information from within the smallest microvessels. The adopted ultrasound contrast agents (UCAs) are coated gas microbubbles. Because they are comparable in size to red blood cells, these microbubbles can flow into the smallest vessels while remaining in the vascular pool (Feinstein 2004). Ultrasound waves induce non-linear bubble oscillations that are exploited by contrast-specific imaging techniques to suppress the signal back-scattered from tissue (de Jong et al. 2000).
Several DCE-US methods have been proposed for angiogenesis detection by assessment of tissue perfusion (Cosgrove and Lassau 2010; Lamuraglia et al. 2010). Typically, quantification is performed by extracting amplitude and time features from time-intensity curves (TICs), which measure the back-scattered acoustic intensity in a selected region of interest (ROI) in the ultrasound image as a function of time. When obtained by the destruction-replenishment technique (Wei et al. 1998), TICs reflect reperfusion after disruption of microbubbles in the image plane (Arditi et al. 2006; Hudson et al. 2009; Krix et al. 2003). Alternatively, the bolus injection technique produces TICs that characterize the passage of an intravenously injected UCA bolus through the image plane (Eckersley et al. 2002; Elie et al. 2007; Strouthos et al. 2010).
In general, however, perfusion quantification has not resulted in reliable angiogenesis detection. One reason for this may be difficult interpretation of TIC features in terms of local perfusion (Tang et al. 2011). Although amplitude-related features are strongly affected by non-linear attenuation and scanner settings (Tang and Eckersley 2006), timing features generally depend on the history of the bolus transport from the injection to the detection site. Another reason for the limited results may be the complicated influence of angiogenesis on microvascular perfusion. While the presence of arteriovenous shunts and a higher microvascular density are expected to increase perfusion, these effects can be countered by an increased interstitial pressure, caused by leakage, and the small diameter and high tortuosity of the newly formed microvessels (Delorme and Knopp 1998; Eberhard et al. 2000; Elie et al. 2007).
Contrast-ultrasound dispersion imaging (CUDI) is an alternative method for detection of the angiogenesis-induced effects on the microvascular architecture (Kuenen et al. 2011; Mischi et al. 2012). CUDI is based on modeling of intravascular UCA transport kinetics as a convective dispersion process. The distribution of transit times is characterized by the local UCA dispersion kinetics resulting from microvascular architectural features such as density, tortuosity and arteriovenous shunting. A dispersion-related parameter, κ, can be estimated by curve fitting of pixel TICs obtained by the bolus injection technique (Kuenen et al. 2011).
Recently, an indirect dispersion analysis by estimation of the spatial similarity among neighbor TICs using coherence analysis was proposed (Mischi et al. 2012). This approach is unique in the sense that it exploits the spatial information. It is different from existing methods because the estimation is inherently local and normalized. Moreover, TIC fitting and isolation of the bolus first pass are not required, resulting in increased robustness of the estimation.
Inspired by the spatiotemporal analysis proposed by Mischi et al. (2012), in this article we propose a new mathematical framework explaining the physical link between dispersion and spatial similarity. Based on a better understanding of the underlying physical processes, we also propose a number of methodological improvements with the aim of improving the reliability and classification performance of the method. In particular, we propose a dedicated spatial filter to prevent local differences in speckle properties from affecting the similarity analysis. Additionally, TIC time windowing is incorporated to make the similarity analysis more specific to TIC shape variations.
A preliminary validation of CUDI was performed with 12 recordings in eight patients who underwent radical prostatectomy at the Academic Medical Center (AMC) University Hospital, Amsterdam, The Netherlands. The histology results were used as ground truth to evaluate the cancer localization performance of the dispersion maps estimated by CUDI. The results were compared with those obtained by estimation of different TIC parameters described in the literature.
Section snippets
Data acquisition and calibration
The DCE-US imaging data were acquired at the AMC University Hospital after the study was approved by the local ethics committee. All patients who participated in this study signed an informed consent.
After intravenous injection of a 2.4-mL UCA bolus (SonoVue, Bracco, Milan, Italy), DCE-US imaging of the prostate was performed with an iU22 ultrasound imaging system (Philips Healthcare, Bothell, WA, USA) and either a C8-4v (six recordings in five patients) or a C10-3v (six recordings in three
Effects of spatial filtering and windowing
The influence of the speckle regularization filter on the similarity analysis was evaluated by comparing the obtained coherence ρ with that obtained by Mischi et al. (2012), that is, after standard Gaussian filtering with σ = 0.25 mm. We observed an average decrease in coherence when the speckle regularization filter was applied, as compared with the use of standard Gaussian filtering. This decrease was generally stronger for greater depths (Fig. 8).
The effect of time windowing on the coherence
Discussion
The current diagnostic limitations that hamper prostate cancer care could be overcome by reliable angiogenesis detection, but the associated microvascular changes are complex and difficult to detect by imaging of perfusion. Recently, we proposed an alternative characterization of the microvascular architecture based on analysis of UCA dispersion kinetics. Dispersion estimation was performed either by curve-fitting of TICs (Kuenen et al. 2011) or by analysis of the spatial TIC similarity (Mischi
Conclusions
Characterization of ultrasound-contrast-agent dispersion by analysis of the similarity among neighbor time-intensity curves has produced promising preliminary results for prostate cancer localization. In this work, we investigated the relation between dispersion and TIC similarity by analysis of the coherence ρ between LDRW TICs. The resulting positive correlation between the dispersion-related TIC parameter κ and TIC similarity was confirmed in the preliminary validation. Dedicated spatial
Acknowledgments
The authors acknowledge the support of the Dutch Technology Foundation (STW).
References (41)
- et al.
Adaptive filtering for reduction of speckle in ultrasonic pulse-echo images
Ultrasonics
(1986) - et al.
Comparison of microscopic vascularity in benign and malignant prostate tissue
Hum Pathol
(1993) - et al.
Detection procedures of ultrasound contrast agents
Ultrasonics
(2000) - et al.
Methodology for quantifying interactions between perfusion evaluated by DCE-US and hypoxia throughout tumor growth
Ultrasound Med Biol
(2007) - et al.
Quantification of flow using ultrasound and microbubbles: A disruption replenishment model based on physical principles
Ultrasound Med Biol
(2009) - et al.
A multivessel model describing replenishment kinetics of ultrasonics contrast agent for quantification of tissue perfusion
Ultrasound Med Biol
(2003) - et al.
Clinical relevance of contrast-enhanced ultrasound in monitoring anti-angiogenic therapy of cancer: Current status and perspectives
Crit Rev Oncol Hematol
(2010) - et al.
Identification of cardiovascular dilution systems by contrast ultrasound
Ultrasound Med Biol
(2007) - et al.
Handling and pathology reporting of radical prostatectomy specimens
Eur Urol
(2003) - et al.
Functional magnetic resonance imaging in prostate cancer
Eur Urol
(2009)