Ultrasound-contrast-agent dispersion and velocity imaging for prostate cancer localization
Graphical abstract
Introduction
Prostate cancer (PCa) is the most frequently diagnosed cancer in men aside from skin cancer, and the second-leading cause of cancer death in men (Society, 2015). Given the significant risk of serious side effects associated with PCa treatment (radical prostatectomy), careful observation (termed active surveillance) instead of immediate treatment is appropriate for many patients that have less aggressive tumors. This approach requires accurate and reliable monitoring techniques. When treatment is necessary, minimally invasive methods such as focal therapy may limit side effects, which in turn requires accurate tumor localization. The current golden standard for prostate cancer diagnosis is transrectal systematic needle biopsies. However, initial biopsies miss nearly a quarter of the clinically significant cancers (Roehl et al., 2002), and provide little information regarding exact tumor locations. Moreover, being an invasive technique, it carries significant risk of infection. This requires hospitalization in up to 6% of the cases (Loeb et al., 2013), becoming even more alarming with increasing resistance to antibiotics. Although transperineal biopsy is emerging as a way to reduce this risk, it is a more complex procedure that requires high grade anesthesia (Chang et al., 2013).
Dynamic Contrast Enhanced Ultrasound (DCE-US) is a minimally invasive diagnostic tool that allows analysis of vascularization, by imaging an intravenously injected microbubble bolus. Of particular interest is the localization of neoangiogenic vascularization associated with tumor growth and metastasis (Folkman, 2002, Brawer, 1996, Weidner, Carroll, Flax, Blumenfeld, Folkman, 1993). In this paper, we aim at characterizing the microvasculature from the obtained indicator-dilution curves (IDCs) using DCE-US; each IDC represents the evolution over time of the ultrasound contrast agent (UCA) concentration in a pixel.
The microvascular structure that originates from tumor driven angiogenic growth is characterized by high microvascular density (MVD), small-diameter vessels that are highly tortuous, chaotic, irregular and have shunts. Ineffective blood flow can lead to hypoxia and deteriorated endothelial wall cells, potentially resulting in extra-vascular leakage and tumor metastases. With the aim of detecting angiogenic microvascularization, DCE-US imaging of hemodynamic features relies on the hypothesis that these features reflect changes in microvasculature associated with angiogenesis. Focusing at increased MVD, time-intensity features related to microvasuclar perfusion have been studied by several researchers (Lueck, Kim, Burns, Martel, 2008, Cosgrove, Lassau, 2010, Wei, Le, Bin, Coggins, Thorpe, Kaul, 2001). However, ultrasound attenuation and scanner settings affect the estimation of local UCA concentration and the resulting amplitude based perfusion parameters. Moreover, increased tortuosity as well as increased flow resistance due to decreasing functional vascular cross-sectional area in neoplastic tissue cause lower tumor perfusion (Gillies et al., 1999), leading to perfusion heterogeneity and making localization of angiogenesis based on perfusion a challenging task. Related to this, intra-tumor vascular heterogeneity has been assessed (Cao et al., 2009), although using DCE-CT instead of DCE-US. To enhance the sensitivity of perfusion imaging, regularized deconvolution of the perfused tissue signals with the feeding-artery signal (referred to as arterial input function) is investigated for DCE-CT and DCE-MRI in Koh et al. (2004).
Alternatively, features linked to UCA bolus dispersion have been proposed (Kuenen, Mischi, Wijkstra, 2011, Kuenen, Saidov, Wijkstra, de La Rosette, Mischi, 2013a), and are instead intended to directly reflect the tortuous and chaotic structure of the tumor vasculature. Although these approaches have shown promise, independent estimation of dispersion and velocity was not possible due to the ambiguity between dispersive and convective processes reflected in the measured IDCs. Hence, so far only dispersion related parameters that represent a combination of dispersion and velocity were obtained, leaving the specific contribution of both components to the flow kinetics unassessed. Furthermore, to achieve a local estimate of the contrast kinetics, a specific spatial UCA bolus concentration profile was assumed.
Instead of modeling the individual measured IDCs, we consider the vascular network as a dynamic linear system or channel, whose impulse response can be locally identified by input-output analysis of IDCs. For this purpose, a Wiener filter is determined, providing an optimal (minimum mean squared error) estimation of the system impulse response. The extraction of the dispersion coefficient, velocity and Péclet number is then facilitated by employing model-based parameter estimation by least squares and Maximum Likelihood approaches.
The analytical details of the measurement model are given in Section 2.1, and an estimator for the Wiener filter is derived in Sections 2.2 and 2.3. Section 2.4 provides a model-based parameter estimator based on Least Squares minimization. Alternatively, Maximum Likelihood estimators are derived in Section 2.5. The data acquisition protocol and the validation methodology are reported in Sections 3 and 4, respectively. The method is then clinically evaluated using a dataset consisting of 61 DCE-US planes, recorded transrectally from 25 patients. A qualitative as well as a quantitative analysis is performed, and the effectiveness of Least Squares and Maximum Likelihood parameter estimators is compared in Section 5. Finally, in Section 6, the results are discussed and conclusions derived.
Section snippets
Measurement model
We consider a ring shaped kernel with an inner and outer radius of 1 mm and 1.5 mm, respectively, as shown in Fig. 1. The dimensions of the kernel were selected based on the speckle-grain size (Kuenen et al., 2013b) and the scale at which early angiogenesis occurs (Brawer, 1996). The kernel should be larger than the system resolution and smaller than the scale at which angiogenesis develops. With the ultrasound system’s axial resolution being approximately 0.3 mm, and the lateral resolution
Data acquisition
Using the methods described in Kuenen et al. (2011), the relation between SonoVue® concentration and acoustic intensity, along with the ultrasound scanner’s compression function were determined and used to estimate the linearized IDCs from the measured acoustic intensity. For SonoVue® concentrations up to 1.0 mg/L, the contrast agent concentration and acoustic intensity were found to be linearly correlated () (Kuenen et al., 2011).
Validation methodology
The performance of the ML and LS estimators across the observations were compared by calculating p-values of the difference between their area under the ROC curve. The required standard errors related to the ROC area are computed according to Hanley and McNeil (1982), and are dependent on the number of independent samples. Taking into account the correlation between pixels within the same ROI, the amount of independent benign and malignant samples is conservatively set to the number of benign
Results
In Fig. 5, the ROC curves for pixel-based classification using the dispersion coefficient, velocity, and Péclet number are given. To compare their performances, the curves when employing ML as well as LS to estimate the model parameters are shown. An overview of these results, including the corresponding PPV, NPV, and the ROC curve areas, is given in Table 1. We show that using ML instead of LS yielded significantly higher ROC curve areas for the estimation of D () and Pe ().
Contributions and strengths
A qualitative comparison of the resulting parametric maps exemplified how the dispersion and velocity maps suggested the presence of angiogenic vasculature by showing dark or bright areas, respectively. The Péclet number map qualitatively displays a higher specificity, with bright areas implying angiogenic activity. Set against the corresponding histology slice, these areas indicated the presence of cancer.
A quantitative analysis showed that the ML parameter estimates outperformed the LS
Conclusions
This paper proposes for the first time a method for the independent estimation of dispersion and velocity of ultrasound contrast agents (UCAs) based on dynamic contrast enhanced ultrasound (DCE-US) imaging. With the aim of localizing prostate cancer (PCa), the developed method enables local characterization of the vascular hemodynamics described by the apparent in-plane dispersion coefficient, velocity, and Péclet number of UCAs. A quantitative analysis on 25 patients revealed that PCa
Acknowledgments
This work is part of the research programme 10769, which is partly financed by the Netherlands Organization for Scientific Research (NWO). The research has also received funding from the European Research Council/ ERC grant agreement no. 280209.
References (50)
- et al.
Red blood cell velocity in nailfold capillaries of man measured by a television microscopy technique
Microvasc. Res.
(1974) - et al.
A comparison of tumor and normal tissue microvascular hematocrits and red cell fluxes in a rat window chamber model
Int. J. Radiat. Oncol. Biol. Phys.
(1993) - et al.
Morphologic and hemodynamic comparison of tumor and healing normal tissue microvasculature
Int. J. Radiat. Oncol. Biol. Phys.
(1989) Role of angiogenesis in tumor growth and metastasis
Seminars in Oncology
(2002)- et al.
Causes and effects of heterogeneous perfusion in tumors
Neoplasia
(1999) Histologic grading of prostate cancer: a perspective
Hum.Pathol.
(1992)- et al.
Detection procedures of ultrasound contrast agents
Ultrasonics
(2000) - et al.
Contrast-ultrasound dispersion imaging for prostate cancer localization by improved spatiotemporal similarity analysis
Ultrasound Med. Biol.
(2013) - et al.
Systematic review of complications of prostate biopsy
Eur.Urol.
(2013) - et al.
Handling and pathology reporting of radical prostatectomy specimens
Eur.Urol.
(2003)
Serial biopsy results in prostate cancer screening study
J. Urol.
3D surface-based registration of ultrasound and histology in prostate cancer imaging
Computerized Med. Imaging Graphics
Prostate-specific antigen
Seminars in Cancer Biology
Capillary blood cell velocity in human skin capillaries located perpendicularly to the skin surface: Measured by a new laser doppler anemometer
Microvasc.Res.
Quantification of renal blood flow with contrast-enhanced ultrasound
J. Am. Coll. Cardiol.
Pattern Recognition and Machine Learning
Quantitative microvessel density: a staging and prognostic marker for human prostatic carcinoma
Cancer
Developing dce-ct to quantify intra-tumor heterogeneity in breast tumors with differing angiogenic phenotype
IEEE Trans. Med. Imaging
Transperineal biopsy of the prostate - is this the future?
Nat. Rev. Urol.
Imaging of perfusion using ultrasound
Eur. J. Nucl. Med. Mol. Imaging
Green’s Functions with Applications
The powerful microbubble: From bench to bedside, from intravascular indicator to therapeutic delivery system, and beyond
Am. J. Physiol. Heart Circulatory Physiol.
Simultaneous determination of the pulmonary and systemic circulation times in man and of a figure related to the cardiac output
Am. J. Physiol. -Legacy Content
The meaning and use of the area under a receiver operating characteristic (roc) curve
Radiology
Endothelial cell death, angiogenesis, and microvascular function after castration in an androgen-dependent tumor: Role of vascular endothelial growth factor
Proc. Natl. Acad. Sci.
Cited by (44)
Contrast-Enhanced Ultrasound Quantification: From Kinetic Modeling to Machine Learning
2020, Ultrasound in Medicine and BiologyCitation Excerpt :Adopting again a ring-shaped kernel with the same rationale as described above, the center TIC is considered the input of the system, while the TICs in the kernel are all possible outputs. Based on the known distance between the ring center and each pixel in the kernel, and including causal contributions only, input–output analysis by model-based system identification enables separate estimation of v and D, from which the Péclet number can also be estimated (van Sloun et al. 2017a). Despite the limitations imposed by a 2-D approach, where only projections along the imaging plane can be visualized, statistical analysis of the obtained velocity vector field allows characterization of the heterogeneous and chaotic nature of angiogenic vasculature.
3-D Multi-parametric Contrast-Enhanced Ultrasound for the Prediction of Prostate Cancer
2019, Ultrasound in Medicine and BiologyCitation Excerpt :Alternatively, the similarity in TIC shape between a voxel and its neighbors can serve as characterization of the dispersion in the vessel architecture connecting those voxels (Kuenen MPJ et al. 2013; Mischi et al. 2012; Schalk et al. 2017). In addition, a system identification approach can be employed to quantify this transition in terms of apparent velocity and dispersion in the underlying convective-dispersion system (van Sloun et al. 2017b). These approaches will be discussed more elaborately in the remaining sections of the materials and methods.
On the Relationship between Dynamic Contrast-Enhanced Ultrasound Parameters and the Underlying Vascular Architecture Extracted from Acoustic Angiography
2019, Ultrasound in Medicine and BiologyCitation Excerpt :Different vascular parameters may contribute to the separate terms of dispersion and flow velocity, whereas we assessed their combination. In this regard, it would also be of interest to apply another analysis to the DCE-US bolus recordings that allows us to separate dispersion and velocity contributions (van Sloun et al. 2017). In this work, dispersion was found to have superior performance in tumor classification compared with perfusion, as previously found for prostate cancer (Kuenen et al. 2013a, 2013b; Mischi et al. 2012).
Multi-layer Convolutional Approach for Lung Cancer Detection using CXR
2023, 2023 4th International Conference on Computing, Mathematics and Engineering Technologies: Sustainable Technologies for Socio-Economic Development, iCoMET 2023