Coherence-based contrast-ultrasound diffusion imaging for
prostate cancer detection
Citation for published version (APA):
Kuenen, M. P. J., Mischi, M., & Wijkstra, H. (2010). Coherence-based contrast-ultrasound diffusion imaging for prostate cancer detection. In Proceedings of the 2010 IEEE Ultrasonics Symposium (IUS), 11-14 October, 2010, San Diego, California (pp. 1936-1939). Institute of Electrical and Electronics Engineers.
https://doi.org/10.1109/ULTSYM.2010.5935893
DOI:
10.1109/ULTSYM.2010.5935893
Document status and date: Published: 01/01/2010
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
Coherence-based Contrast Ultrasound Diffusion
Imaging for Prostate Cancer Detection
M. P. J. Kuenen and M. Mischi
Eindhoven University of Technology Eindhoven, the Netherlands Email: m.p.j.kuenen@tue.nl
H. Wijkstra
AMC University Hospital Amsterdam, the Netherlands
Abstract—Prostate cancer is the most common form of cancer
in men in western countries. The use of efficient focal therapies is currently hampered by limitations in early prostate cancer detec-tion. With limited success, several quantitative ultrasound perfu-sion imaging methods have aimed at detection of microvascular changes associated to cancer growth. Alternatively, we recently introduced contrast ultrasound diffusion imaging, hypothesizing that these complex microvascular changes are better reflected by diffusion than by perfusion. In this paper we introduce the analysis of spatial similarity as an indirect estimation of diffusion. The passage of an intravenously injected contrast-agent bolus is recorded by transrectal ultrasound imaging, thereby measuring indicator dilution curves with a pixel resolution. The spatial similarity among these curves, within a kernel determined by the ultrasound scanner resolution, is estimated using coherence analysis. The coherence images generated from four patients were compared with histology data on a pixel basis. The results show a receiver operating characteristic curve area of 0.91, higher than that of any perfusion-related parameter. Although a method optimization and an extensive validation are required, these results confirm the promising value of contrast ultrasound diffusion imaging for prostate cancer detection.
I. INTRODUCTION
Prostate cancer is the most common form of cancer in men in the USA, accounting for 28% and 11% of all cancer diag-noses and deaths, respectively [1]. Recently, various efficient focal treatments have become available [2], but their use is hampered by limitations in early prostate cancer detection.
Currently, prostate cancer diagnosis is based on one or more systematic biopsy investigations, which are performed if the suspect of prostate cancer is raised by noninvasive methods such as the PSA blood test [3], [4]. A biopsy investigation commonly involves taking 6 to 12 geometrically distributed tissue samples from the prostate with a hollow needle [3]. In retrospective, only 24% of the patients undergoing biopsy investigations are diagnosed with prostate cancer [4]. This is mainly due to the limited specificity of noninvasive methods. In addition, the lack of localization makes accurate biopsy targeting not feasible.
As a result, new noninvasive methods are necessary to improve early prostate cancer detection. In particular, imaging methods may provide cancer localization to improve biopsy targeting and focal treatment guidance. In this perspective, a
This work was supported by the Dutch Organization for Scientific Research (NWO) and the Technology Foundation (STW).
key prognostic indicator for prostate cancer is angiogenesis, i.e., the formation of a microvascular network that is charac-terized by an increased microvessel density [5]. Angiogenesis correlates to cancer aggressiveness and is required for cancer to grow beyond 1 mm3 [5].
Several methods have aimed at detection of angiogene-sis by quantitative transrectal ultrasound (TRUS) imaging of tissue perfusion, i.e., blood flow per tissue volume [3]. These methods use ultrasound contrast agents (UCAs) to detect microvascular flow [3], [6]. UCAs are dispersions of gas microbubbles that backscatter acoustic energy when invested by ultrasound waves [7]. High acoustic pressures cause microbubble disruption. This effect, which distorts con-ventional Doppler flow measurements [6], is exploited by the destruction-replenishment technique [8], [9]. A different technique uses low pressures for dynamic TRUS imaging after an intravenous UCA bolus injection. However, only qualitative results are reported [10], [11].
Up until now, despite improvements in biopsy targeting, quantitative ultrasound perfusion imaging has not yet proven sufficiently reliable to replace systematic biopsies [6], [9]. Apart from limitations in flow sensitivity, this may be due to the complex and contradictory effects of angiogenesis on perfusion [12]. At first glance, a higher number of microves-sels would result in an increased perfusion, but this may be counterbalanced by e.g. an increased interstitial pressure [12]. Recently, we introduced contrast ultrasound diffusion imag-ing as an alternative to perfusion imagimag-ing [13]. In this context, diffusion describes the intravascular UCA spreading by both apparent diffusion and convective dispersion, due to multipath trajectories [14]. In fact, we consider the microvascular archi-tecture as a distributed network, in which flow can be modeled as flow through porous media [15]. The structural charac-teristics of such networks can be characterized by diffusion [14]. Therefore, we hypothesize that angiogenesis-induced microvascular changes are correlated with UCA diffusion.
In a previous preliminary study using the bolus injection technique, we showed that a diffusion-related parameter dif-ferentiated between cancerous and healthy tissue better than perfusion-related parameters [13]. By modeling the UCA transport by the convective-diffusion equation, we obtained a formalization that enables the extraction of a local diffusion-related parameter from indicator dilution curves (IDCs) that
Fig. 1. Contrast-enhanced TRUS of the prostate, featuring the power modulation (left) and fundamental (right) imaging modes.
can be measured at each pixel.
In this paper, we introduce a new method for an indirect estimation of UCA diffusion, based on coherence analysis. Rather than focusing on each IDC individually, this method exploits the spatial effects of diffusion by considering the simi-larity among spatially adjacent IDCs as an indirect measure for diffusion. A coherence-based similarity analysis, independent of IDC arrival times, is adopted for UCA diffusion imaging.
II. MATERIALS ANDMETHODS
A. Data acquisition and calibration
The data acquisition was performed at the AMC Uni-versity Hospital (Amsterdam, The Netherlands). A 2.4 mL SonoVue bolus was injected in an arm vein. TRUS imagingR
was then performed with an iU22 ultrasound scanner (Philips Healthcare, Bothell, WA) equipped with a C8-4v probe. Power modulation mode was adopted for contrast-specific imaging [16]. The ultrasound frequency was 3.5 MHz and the mechan-ical index was 0.06. The acquired ultrasound image sequences were stored in DICOM (Digital Imaging and Communications in Medicine) format, suitable for our Matlab-based analysis.R
An example image is shown in Fig. 1.
To map the image gray level to the UCA concentration, it is necessary to compensate the gray scale mapping and the compression function that are performed by the scanner. For the given compression setting, we estimated the compression function by comparing Q-Lab acoustic quantification resultsR
in single-pixel regions of interest (ROIs) to the image gray level. This comparison showed a linear relation (R2 = 0.99) between the acoustic intensity in dB and the gray level, which implies a logarithmic compression function. We can recon-struct the backscattered acoustic intensity by compensating for the gray mapping and this compression function. Since the backscattered acoustic intensity relates linearly to the UCA concentration for low concentrations and low mechanical
index [17], we obtain a linear measure for the UCA concen-tration [13].
B. Diffusion and spatial similarity
We illustrate the effects of diffusion on the spatial IDC similarity using the Local Density Random Walk (LDRW) model [13], [18], [19]. This model is a solution of the convective-diffusion equation in a straight tube, in which a carrier fluid flows at constant velocity v. For boundary conditions that imply UCA mass conservation and a rapid UCA bolus injection, the UCA concentration dynamicsC(z, t) is given as C(z, t) = m A√4πDtexp −(z − vt)4Dt 2 . (1)
In (1),t and z represent the time since the bolus injection and the distance between the injection and measurement positions, respectively. The parameters D, m and A represent the UCA diffusion coefficient, the injected UCA dose mass, and the tube section, respectively.
To show the relation between IDCs measured at different positions, we derive C(z + Δz, t) as C(z + Δz, t) = C(z, t) exp v −z t Δz − Δz2 2t 2D . (2)
From (2), we observe that for largeD, C(z, t) ≈ C(z+Δz, t). On the other hand, for smallD, the difference between C(z, t) and C(z + Δz, t) increases. This example shows the link between the similarity of spatially adjacent IDCs and the UCA diffusion coefficient.
Compared to our previous method [13], an indirect analysis of diffusion by measuring the spatial IDC similarity offers several advantages. The spatial IDC similarity characterizes lo-cal microvascular characteristics without making assumptions about the spatial distribution of the UCA bolus flowing into the prostate circulation. Also, no IDC fitting is required for this approach. Furthermore, isolation of the first bolus passage is not required, reducing the signal-to-noise ratio requirements.
A straightforward similarity measure is provided by the correlation coefficient between IDCs, which is however in-fluenced by differences in arrival times. These differences are rather a measure of perfusion than diffusion.
Alternatively, we have decided to compare the IDCs in the frequency domain by coherence analysis, which focuses on the IDC magnitude spectra. Arrival time differences, which are incorporated in the phase, have no influence on such analysis. In addition, the frequency range covering the UCA transport dynamics can be easily isolated.
C. Coherence analysis
For each pixel, the IDC spectrum is compared to the spectra of neighbor pixels. A spatial kernel determines which pixels are used for this comparison. The geometry of this kernel is based on the TRUS spatial resolution and the scale of microvascular networks for which diffusion can be estimated.
1.0 mm 0.5 mm
Central pixel Compared pixel Other pixel
Fig. 2. Kernel adopted for spatial similarity analysis.
In this method, we only considered the axial resolution, which is determined by the transmitted pulse length. The transmitted pulses have a frequency of 3.5 MHz and have an effective length of two cycles, which provides an axial resolution of 0.43 mm.
The microvascular scale of interest is given by the smallest tumor size for which angiogenesis can be detected. Angiogen-esis is required for tumors to grow beyond 1 mm3 [5].
Based on these constraints, the proposed kernel implements a ring shape, as indicated by Fig. 2. To prevent the ultrasound resolution from influencing the result, pixels within a range of 0.5 mm are not compared. The outer radius of 1 mm ensures a sufficient resolution for the microvascular scale of interest.
The analysis is restricted to the frequency range covering the UCA transport dynamics. Because this dynamics is limited to frequencies up to 0.5 Hz [13], frequencies above 0.5 Hz are discarded. The DC component is also discarded, as it is strongly affected by scanner settings such as the gain.
Coherence analysis is then performed as follows. First, the IDC magnitude spectrum is computed at all pixels, for the specified frequency range. Then, the correlation coefficient is computed between the spectrum at the central pixel and the spectrum at each of the pixels within the ring. These coefficients are averaged to obtain a coherence measure.
This procedure is repeated for all pixels covering the prostate, resulting in a coherence image. Prior to display, the coherence image is filtered using a Gaussian kernel with a standard deviation of 0.5 mm. This postprocessing step aims at emphasizing diffusion at the scale where microvascular networks can be associated with the presence of cancer. An example of a result image is shown in Fig. 3.
D. In-vivo validation
We evaluated the cancer localization capability of coherence imaging by comparing the coherence images from four pa-tients with histology data, obtained after radical prostatectomy at the AMC University Hospital (Amsterdam, The
Nether-TABLE I
SENSITIVITY,SPECIFICITY,ANDROCCURVE AREA ON PIXEL BASIS OF THEIDCCOHERENCE AS WELL AS SEVERAL HEMODYNAMIC
PARAMETERS EXTRACTED FROM FITTEDIDCS.
Parameter Cancerous if Sensitivity Specificity ROC curve area
Coherence ≥ 0.484 80.3 % 86.3 % 0.911 κ ≥ 0.856 s−1 81.6 % 78.3 % 0.882 AUC ≥ 904 62.2 % 87.7 % 0.779 PV ≥ 60.0 75.5 % 94.1 % 0.851 PT ≤ 29.5 s 80.7 % 51.7 % 0.693 AT ≤ 18.7 s 78.5 % 45.3 % 0.625 FWHM ≤ 14.6 s 87.3 % 67.2 % 0.828 WIT ≤ 9.62 s 85.3 % 60.8 % 0.775
lands). After cutting the prostate in slices of 4-mm thickness, a pathologist marked the presence of cancer [20]. We selected the slice corresponding to the ultrasound imaging plane and compared the results.
For each prostate, we selected two ROIs larger than 0.5 cm2, representing healthy and cancerous tissue. An example is shown in Fig. 3. We measured the statistical mean and variance of coherence for each class (healthy and cancerous tissue) on a pixel basis. We then used Bayes inference to obtain the optimal threshold for classification, and we derived the optimal sensitivity and specificity [21]. In addition, we evaluated the receiver operating characteristic (ROC) curve. For comparison, this procedure was also performed for the local diffusion parameter κ [13] and for several perfusion-related IDC parameters [22], [23] that were obtained after LDRW model fitting.
III. RESULTS
A higher IDC coherence was observed in all patients in the presence of cancer. In the four datasets considered in this study, we obtained an average coherence of 0.601 ± 0.137 in ROIs representing cancerous tissue, and 0.366± 0.108 in ROIs representing healthy tissue. The optimal sensitivity and specificity, as well as the ROC curve area for cancer detection on a pixel basis are reported in Table I.
This procedure was repeated for several IDC parameters that have been proposed in the literature [13], [22], [23]. The considered parameters are the area under the curve (AUC), the IDC peak value (PV), the IDC peak time (PT), the contrast appearance time (AT), the IDC full-width at half maximum (FWHM) and the wash-in time (WIT = PT-AT).
IV. DISCUSSION ANDCONCLUSIONS
In this paper, we present a new method for indirect estima-tion of diffusion based on the spatial similarity among IDCs. The similarity is estimated by an IDC coherence analysis within a spatial kernel that is based on the ultrasound scanner resolution as well as the size of clinically significant cancer.
In comparison to our previous method [13], coherence-based diffusion imaging offers several advantages. Coherence analysis uses spatial information and requires no model fitting. Therefore, no assumptions are made about the IDC shape or
0.10 0.90
Fig. 3. Coherence image (left), fundamental ultrasound image (center), and histology (right). On the left, the average spatial IDC similarity, measured by coherence is overlaid on the ultrasound image by color coding. The manually selected white contour represents the prostate boundary. The red and green polygons in the central image represent the adopted ROIs for cancerous and healthy tissue, respectively.
the UCA spatial distribution entering the prostate circulation. Moreover, there is no need for isolation of the bolus first pass. The preliminary validation of this method shows slightly superior results than those obtained by our previous contrast-ultrasound diffusion imaging method [13]. In four patients, both diffusion-related parameters show a better correspon-dence to the histology than any perfusion-related IDC parame-ter. A more extensive and improved validation, aimed at a more accurate match between ultrasound and histology images, is however necessary. To this end, three-dimensional contrast-enhanced TRUS may offer new possibilities in the future. This would enable the investigation of the entire prostate with a single bolus injection, and improve the spatiotemporal analysis of UCA diffusion.
In the future, alternative similarity measures will also be investigated. In addition, the nonuniform ultrasound resolution will be taken into consideration in the kernel design.
REFERENCES
[1] American Cancer Society, “Cancer facts & figures,” Atlanta, 2010. [2] T. J. Polascik and V. Mouraviev, “Focal therapy for prostate cancer,”
Curr. Opin. Urol., vol. 18, pp. 269–274, 2008.
[3] H. Wijkstra, M. H. Wink, and J. J. M. C. H. de la Rosette, “Contrast specific imaging in the detection and localization of prostate cancer,”
World J. Urology, vol. 22, pp. 346–350, 2004.
[4] F. H. Schr¨oder, J. Hugosson, M. J. Roobol et al., “Screening and prostate-cancer mortality in a randomized European study,” N. Engl.
J. Med., vol. 360, no. 13, pp. 1320–1328, 2009.
[5] M. K. Brawer, “Quantitative microvessel density. A staging and prog-nostic marker for human prostatic carcinoma,” Cancer, vol. 78, no. 2, pp. 345–349, 1996.
[6] E. J. Halpern, J. R. Ramey, S. E. Strup et al., “Detection of prostate car-cinoma with contrast-enhanced sonography using intermittent harmonic imaging,” Cancer, vol. 104, pp. 2373–2383, 2005.
[7] S. B. Feinstein, “The powerful microbubble: from bench to bedside, from intravascular indicator to therapeutic delivery system, and beyond,” Am.
J. Physiol. Heart Circ. Physiol., vol. 287, no. 2, pp. H450–H457, 2004.
[8] K. Wei, A. R. Jayaweera, S. Firoozan et al., “Quantification of myocar-dial blood flow with ultrasound-induced destruction of microbubbles administered as a constant venous infusion,” Circulation, vol. 97, pp. 473–483, 1998.
[9] R. A. Linden, E. J. Trabulsi, F. Forsberg et al., “Contrast enhanced ultrasound flash replenishment method for directed prostate biopsies,”
J. Urol., vol. 178, pp. 2354–2358, 2007.
[10] M. H. Wink, J. J. M. C. H. de la Rosette, C. A. Grimbergen, and H. Wijkstra, “Transrectal contrast enhanced ultrasound for diagnosis of prostate cancer,” World J. Urology, vol. 25, pp. 367–373, 2007. [11] F. Aigner, L. Pallwein, M. Mitterberger et al., “Contrast-enhanced
ultrasonography using cadence-contrast pulse sequencing technology for targeted biopsy of the prostate,” B. J. U. Int., vol. 103, pp. 458–463, 2008.
[12] S. Delorme and M. V. Knopp, “Non-invasive vascular imaging: assessing tumour vascularity,” Eur. Radiol., vol. 8, pp. 517–527, 1998.
[13] M. Mischi, M. P. J. Kuenen, H. Wijkstra, A. J. M. Hendrikx, and H. H. M. Korsten, “Prostrate cancer localization by contrast-ultrasound diffusion imaging,” in 2009 IEEE Ultrasonics Symposium. IEEE UFFC Society, 2009.
[14] R. J. M. de Wiest, Flow through porous media. London: Academic Press, 1969.
[15] S. J. Chapman, R. J. Shipley, and R. Jawad, “Multiscale modeling of fluid transport in tumors,” Bull. Math. Biol., vol. 70, pp. 2334–2357, 2008.
[16] N. de Jong, P. J. A. Frinking, A. Bouakaz, and F. J. ten Cate, “Detection procedures of ultrasound contrast agents,” Ultrasonics, vol. 38, no. 1–8, pp. 87–92, 2000.
[17] M. Mischi, A. A. C. M. Kalker, and H. H. M. Korsten, “Videodensit-ometric methods for cardiac output measurements,” EURASIP J. Appl.
Sign. Proc., vol. 5, pp. 479–489, 2003.
[18] C. W. Sheppard, Basic principles of the tracer method. New York: John Wiley and Sons, 1962.
[19] K. H. Norwich and S. Zelin, “The dispersion of indicator in the cardio-pulmonary system,” Bull. Math. Biophys., vol. 32, pp. 25–43, 1970. [20] R. Montironi, T. van der Kwast, L. Boccon-Gibod, A. V. Bono, and
L. Boccon-Gibod, “Handling and pathology reporting of radical prosta-tectomy specimens,” Eur. Urol., vol. 44, pp. 626–636, 2003.
[21] C. M. Bishop, Pattern Recognition and Machine Learning. Springer, 2006.
[22] N. Elie, A. Kaliski, P. P´eronneau et al., “Methodology for quantifying interactions between perfusion evaluated by DCE-US and hypoxia throughout tumor growth,” Ultrasound in Med. & Biol., vol. 33, no. 4, pp. 549–560, 2007.
[23] C. Strouthos, M. Lampaskis, V. Sboros, A. McNeilly, and M. Averkiou, “Indicator dilution models for the quantification of microvascular blood flow with bolus administration of ultrasound contrast agents,” IEEE
Trans. Ultrason., Ferroelect., Freq. Contr., vol. 57, no. 6, pp. 1296–
1310, 2010.
