2D Contrast-Ultrasound Dispersion Imaging of Angiogenesis in Adenomyosis: First Experimental Measurements

2D Contrast-Ultrasound Dispersion Imaging of Angiogenesis in Adenomyosis: First Experimental Measurements Ferenc Igor Kandi Biomedical Diagnostics Lab Eindhoven University of Technology Eindhoven, the Netherlands [email protected] Catarina Dinis Fernandes Biomedical Diagnostics Lab Eindhoven University of Technology Eindhoven, the Netherlands [email protected] Simona Turco Biomedical Diagnostics Lab Eindhoven University of Technology Eindhoven, the Netherlands [email protected] Eva de Bock Gynaecology Amsterdam University Medical Center Amsterdam, the Netherlands [email protected] Lynda Juffermans Gynaecology Amsterdam University Medical Center Amsterdam, the Netherlands [email protected] Judith Huirne Gynaecology Amsterdam University Medical Center Amsterdam, the Netherlands [email protected] Massimo Mischi Biomedical Diagnostics Lab Eindhoven University of Technology Eindhoven, the Netherlands [email protected] Abstract—Adenomyosis, a benign yet serious uterine condition, is highly prevalent but difficult to diagnose and assess using conventional B-mode ultrasound imaging. As the formation of adenomyosis involves angiogenesis and influences local microvas- cular dispersion, monitoring microvascularity could improve adenomyosis diagnostics. Hence, contrast-enhanced ultrasound (CEUS), a non-invasive imaging technique, is explored as an alternative to B-mode as it allows visualizing microvascularity. The goal is to investigate the potential of characterizing adeno- myosis by quantifying surrogate measures of contrast dispersion, using the contrast-ultrasound dispersion imaging (CUDI) frame- work based on uterine CEUS. CUDI models the passage of an ultrasound-contrast-agent (UCA) bolus through the uterine tissue as a convective dispersion process and expresses UCA dispersion kinetics through temporal fitting ( κ) and spatiotemporal similar- ity analysis. Before applying CUDI on clinical data, the speckle size is regularized to yield isotropy and depth-independence. In addition, the relationship between measured acoustic inten- sities and underlying UCA concentrations is determined to be linear ( R2=0.94) for SonoVue T M UCA concentrations of 0.15- 1.00 mg/L, allowing for the direct application of the proposed convective-dispersion modeling. After regularization, 2D CEUS acquisitions of one adenomyotic and one healthy uterus under- went CUDI analysis. The adenomyotic myometrium presented heterogeneously higher values of κ and spatiotemporal similarity, for the healthy uterus the values were homogeneous and lower. Since κ and spatiotemporal similarity are inversely proportional to the level of local dispersion, the results indicate the presence of angiogenesis in adenomyotic tissue, supported by the observations of heterogeneously decreased local dispersion. Funded by NWO TTW, project 18482 Index Terms —adenomyosis, uterus, contrast-enhanced ultra- sound, contrast-ultrasound dispersion imaging, temporal fitting, spatiotemporal similarity I. I NTRODUCTION Benign uterine conditions, particularly adenomyosis, are greatly prevalent among women [1]. While these conditions are classified as benign, they can cause heavy discomfort, abnormal bleeding and fertility problems. Prevalence esti- mations vary widely, potentially stemming from a current lack of robust, objective and quantitative diagnostic methods [2]. Current diagnostic protocols rely on B-mode ultrasound and magnetic resonance imaging (MRI) to highlight uter- ine pathologies, but often prove inadequate especially for adenomyosis due to its diffuse nature and wide anatomical variability [2]. This highlights the necessity for an objective diagnostic method which can adapt to the widely varying characteristics of benign uterine conditions, describing them both quantitatively and qualitatively. Adenomyosis has been associated with abnormal vascularization, with angiogenesis being a particularly important driving mechanism supporting disease progression and development [3]. Since conventional B-mode ultrasound does not provide sufficient differential diagnostic power in terms of imaging vascularity, 2D contrast- enhanced ultrasound (CEUS) is employed to visualize (mi- cro)vasculature and therewith angiogenesis [2]. CEUS is an ultrasound imaging technique where an ultrasound contrast agent (UCA) is intravenously administered and the (mostly) This full-text paper was peer-reviewed at the direction of IEEE Instrumentation and Measurement Society prior to the acceptance and publication. 2024 IEEE International Symposium on Medical Measurements and Applications (MeMeA) | 979-8-3503-0799-3/24/$31.00 ©2024 IEEE | DOI: 10.1109/MEMEA60663.2024.10596726 Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply. nonlinear backscatter of the UCA is isolated [4]. Since normal tissue backscatter is mostly linear, the UCA signal is differ- entiable. In CEUS, adenomyosis has a diffuse appearance, presenting both hyper and hypo-enhanced spots. Contrast-ultrasound dispersion imaging (CUDI) [4] provides a powerful framework for the quantification of angiogenesis based on CEUS. Within this framework, contrast time-intensity curve fitting and analysis can be used to characterize contrast dispersion and flow, based on tracking the wash-in and wash- out of a bolus of UCA. In CUDI, the pixel intensities are tracked over time and a convective dispersion model is fitted to this data. Additionally, by comparing the intensities of a pixel with its surroundings in time and frequency domain and through statistical analysis, CUDI can translate this spatiotem- poral similarity analysis into dispersion-related metrics [5] [6], where a decrease in dispersion indicates a more tortuous and dense formation of microvascularity, e.g., angiogenesis. Adenomyosis spreads heterogeneously throughout the uterine myometrium and angiogenesis takes place to account for the locally increased metabolic demands, decreasing the local dispersion. Therefore, the aim is to investigate the use of CUDI for the characterization of adenomyosis, since the hypothesis is that the heterogeneity and decreased local dispersion of the adenomyotic vascularity will be reflected in the temporal fitting and spatiotemporal similarity analyses. II. M ETHODOLOGY Although CUDI has been extensively used in the prostate [5] [7], this is the first time that it is translated to the uterus. Therefore, it is required to perform calibrations specific to the investigated organ and employed ultrasound scanner. For B- mode and CEUS image acquisitions, a HERA W10 (Samsung Medison, Republic of Korea) ultrasound system with an EV2- 10A endovaginal transducer is used. A. Image preprocessing: speckle size regularization Speckles can be defined as the granular texture on the images, caused by interference of the backscatter signals. Prior to any quantitative analysis of the acquired CEUS data, speckle size calibration steps are required, relying on in-vitro experimental measurements. 1) Rationale: The speckle size is anisotropic throughout the acquisition field of view, which is an effect of both the diverging scanlines (native to using convex transducer arrays) and ultrasound propagation physics. As the scan lines diverge, the speckles become increasingly stretched in the lateral direction for increasing depth. For spatiotemporal similarity analysis, CUDI uses an annular kernel and therefore assumes an isotropic speckle size. To avoid inefficient computations in the following analyses, i.e., adapting the annular kernel to take this anisotropy into account, the speckle size is regularized to ensure isotropy in the field of view and the use of one single kernel size for the full image [5]. 2) Measurement set-up & protocol: To define the parame- ters for regularization, the current speckle size distribution is determined by imaging a known concentration of free-floating UCA micro-bubbles using 2D CEUS. The used UCA is SonoVueT M (Bracco Suisse SA, Geneva, Switzerland), which are micro-bubbles consisting of an inert sulphur hexafluoride gas core encapsulated by a phospholipid shell. A large (60- L) body of degassed water, contained in an acoustic-padded polypropylene plastic container, is used as the imaging vessel. The UCA (SonoVue T M) is prepared directly in the imaging vessel to yield the concentration of 0.05 mg/L, as this prede- termined concentration enables measurements of individually discernible micro-bubbles. The probe is submerged such that the transducer is just below water level and a 20-second 2D CEUS recording (gain 30 dB, dynamic range 45 dB, mechanical index 0.1, 120-degree field of view at 12-cm depth) is acquired. 3) Processing: Working under the assumption that speckles have a Gaussian shape, contour points of the speckles (here- with also the spatial resolution) are estimated though the full- width at half-maximum (FWHM) of the 2D autocovariance function (15x15 pixel kernel) [5]. Through these contour points, ellipsoids are fitted using a least-squares approach, followed by extraction of the axial and lateral axes from the ellipsoids. The dimensions of these axial and lateral axes can then be coupled to the axial and lateral dimensions of the speckles in the unregularized image. The speckle size is regularized to be isotropic and equal to the axial size of a speckle in the unregularized image at the depth of interest. Assuming a noise-to-signal ratio (NSR) of 1−3 and a σ of 0.03 cm, Wiener deconvolution is employed to enforce an isotropic speckle size [5], followed by low-pass filtering using an isotropic Gaussian kernel for high-spatial-frequency noise reduction and to obtain a desired resolution [5]. B. Image preprocessing: concentration-intensity calibration For visualizing backscatter ultrasound waves, the dynamic range is compressed to quantize the signal, typically by using a logarithm-like function. This compression function therefore relates acoustic intensity to quantized grey level (in this case through 8-bit greyscale). 1) Rationale: Having access to this function means that UCA concentrations can be assessed in terms of acoustic intensity (and not quantization levels), which is assumed to be linearly related to lower concentrations of UCA [8], since higher concentrations of UCA will cause self-attenuation resulting in lower observed intensities. Establishing and veri- fying this relationship is essential for the correct application of pharmacokinetic modeling. 2) Measurement set-up & protocol: The relationship be- tween acoustic intensity and UCA concentration is investigated by imaging a series of phantoms filled with known (yet different) UCA concentrations using 2D CEUS. A large (60- L) body of degassed water, contained in an acoustic-padded polypropylene plastic container, is again used as the imaging vessel. SonoVueT M is prepared to yield the following concen- trations: 0.15; 0.30; 0.50; 1.00; 1.25; 1.80; 2.50; 5.00; 7.50; 10.00; 12.50; 15.00 and 20.00 mg/L. The rationale behind these concentrations is in that there is no prior knowledge on Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply. which UCA concentrations can be expected within the uterus, hence a wide range of concentrations should be measured (focusing on lower concentrations) and taken into account when relating acoustic intensity and UCA concentration. Then, phantoms (condoms) are individually filled with 100 mL of each of the prepared UCA concentrations, i.e., one phantom per concentration to be measured. The probe is submerged such that the transducer is just below water level and 20- second 2D CEUS recordings (gain 30 dB, dynamic range 45 dB, mechanical index 0.1, 120-degree field of view at 6-cm depth) are acquired for each phantom, moving the phantom through the field of view while the probe is acquiring (to account for potential concentration inhomogeneities within the phantom). 3) Processing: The means of the measured acoustic inten- sities per phantom are calculated and then visualized against their respective UCA concentrations. A linear least-squared error fit is performed through this data and the goodness of fit is assessed through the R2 value. This fit yields the intended concentration-intensity relationship [8]. C. Clinical data The data of 1 adult female participant with clinically confirmed diffuse adenomyosis and 1 healthy adult female (both participants provided written informed consent) was acquired within a pilot study of the UteroVue study (NWO TTW 18482) at the outpatient gynaecology clinic of the Am- sterdam University Medical Center (Amsterdam, the Nether- lands). This study has been approved by the Medical Ethics Committee of the Amsterdam University Medical Center (NL83391.018.23). Sagittal 2D CEUS recordings of the uteri were performed during peripheral intravenous bolus adminis- trations of SonoVueT M UCA, with the probe fixed in a custom probe holder for stability. D. CUDI: time-intensity curve fitting of clinical data To demonstrate the application of the proposed image optimization and calibration in clinical data, the uterine 2D CEUS recordings are quantitatively analyzed using CUDI. The wash-in and wash-out of UCA can be expressed in a time- intensity curve (TIC), describing UCA transport kinetics. To fit a curve to the acoustic intensities of the backscatter data, first a preliminary fit to the data is performed with the sole purpose of identifying the first pass of the UCA through the region of interest (ROI). Then, a solution to the convective dispersion equation in one dimension, a modified local density random walk model (mLDRW), is used to fit a curve to the time-intensity data based on the first pass. The mLDRW model can be described as [4] C(t) = α s κ 2π(t − t0) e(− κ(t−t0 −µ)2 2(t−t0 ) ), (1) where C(t) is the UCA concentration after an intravenous bolus injection as a function of time t, α is a unitless scale metric related to blood flow and UCA dose representing the area under the C(t), t0 refers to the theoretical injection time, µ is the mean transit time of a micro-bubble between the injection site and the location of interest. κ is the local, dispersion-related parameter governing the shape of C(t) relating flow velocity and dispersion to local microvascular architectures. The value of κ can be expressed as κ = v2 2D [1/s], (2) where κ relates flow velocity v and dispersion D [8], which cannot be measured directly; hence, κ is estimated as a surrogate. The following two cases are evident: lower values of κ indicate an increase in dispersion and/or a decrease in velocity, with the TIC being more skewed, while higher values of κ indicate a decrease in dispersion and/or an increase in velocity, with the TIC being more symmetric. As described earlier, CUDI analyzes the full wash-in and wash-out of a UCA bolus, translating this to local dispersion characteristics which are expressed through parametric maps [8]. In turn, local dispersion can be linked to properties of angiogenesis such as microvascular tortuosity and density. Alongside κ, CUDI temporal fitting analysis describes local (pixel-wise) TICs in terms of appearance time of UCA, mean transit time, time to peak, peak intensity and FWHM (full curve width at half of the maximum peak intensity). E. CUDI: spatiotemporal similarity analysis Since the spatiotemporal similarity between neighboring TICs is governed by local dispersion, spatiotemporal com- parison of neighboring TICs yields surrogate estimators of dispersion [4]. To perform this comparison, an annular kernel around a central pixel is used to avoid a flow-directional bias [5]. The dimensions of this kernel are determined to cover the dimensions assumed for early formations of angiogenesis: the inner diameter is determined by the resolution of the system to be 1 mm, the outer diameter is the angiogenesis- informed value of 2.5 mm. Then, the TIC of the central pixel is compared to TICs of pixels in the kernel and their average similarity is expressed as temporal correlation, spectral coherence, and mutual information. This process is repeated for all pixels in the ROI and the results are reported in the form of parametric maps. All the mentioned metrics of spatiotemporal similarity are inversely proportional to the level of local dispersion; a higher similarity indicates less dispersion and vice-versa [4]. 1) Temporal correlation: When assessing spatiotemporal similarity by correlation, the first step is to time-window the signal. Hereafter, the central pixel’s TIC is correlated with each TIC in the kernel and these values are averaged to yield the average temporal correlation. This process is repeated for each pixel in the ROI to construct the correlation parametric map [5]. 2) Spectral coherence: To avoid time-dependencies due to varying arrival times of the bolus (phase information), spectral coherence regards the TIC signals in frequency (Fourier) domain. The frequency range describing the UCA transport kinetics is derived from prostate applications of CUDI [5] Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply. and within this range the amplitudes of the frequency spectra (no phase) are calculated. From here, the amplitudes of the frequency spectra of the central pixel TIC and the TICs of pixels in the kernel are correlated and averaged. This process is repeated for each pixel in the ROI to construct the coherence parametric map [5]. 3) Mutual information: Exploring potential non-linear sim- ilarities between pixels, mutual information analysis samples the central pixel’s TIC and the TICs of pixels in the kernel and regards those sampled signals as random variables. The statistical dependence between those is then measured in terms of mutual information. This process is repeated for each pixel in the ROI to construct the mutual information parametric map [6]. III. R ESULTS A. Speckle size regularization A resulting 2D autocovariance plot can be observed in Fig. 1, alongside its location as denoted on the CEUS image. The axial and lateral speckle dimensions prior to regularization can be observed in Fig. 2 in blue, where the anisotropy is notable through the consistent sub-millimeter axial size and the increasing lateral size. The determined axial and lateral dimensions of the FWHM of the 2D autocovariance can be expressed through: F W HMaxial = 0.0631 − 0.0012 × depth, (3) F W HMlateral = 0.0539 + 0.116 × depth. (4) The resulting speckle dimensions after regularization can be visually appreciated in Fig. 2 in yellow, where now both the axial and the lateral dimensions are consistently around 1 mm and the lateral speckle size shows reduced depth-dependency. B. Concentration-intensity calibration The linear fit between the averaged acoustic intensity and the lower concentrations (0.15; 0.30; 0.50 & 1.00 mg/L) is shown in Fig. 3, reporting an R2 = 0.94. For concentrations up to 1.00 mg/L, this linear relationship between UCA con- centrations C [mg/L] and acoustic intensities I [a.u.] can be expressed as I = 889.17 × C + 418.46. (5) Here, the sensitivity is defined by 889.17 and the background acoustic intensity or noise floor, stemming from the surround- ings, equals 418.46 arbitrary units (a.u.) [8]. C. CUDI After applying the proposed regularization and calibration, CUDI temporal fitting and spatiotemporal similarity analysis is performed on the 2D CEUS acquisitions of the adenomyotic uterus and the healthy uterus. Examples of the temporal fitting (κ) and spatiotemporal similarity parametric maps for the correlation, coherence and mutual information of the myometrium (muscular tissue layer) of the adenomyotic uterus are observable in Fig. 4. Idem, the equivalents for the healthy uterus are visualized in Fig. 5. Comparing the uteri, Fig.6 yields the distributions of the CUDI parameters for the same uteri as are observable in Figs. 4-5. IV. D ISCUSSIONS A. Speckle size regularization From observing both the CEUS image and the autoco- variance from the denoted region in Fig. 1, it is evident that the speckle sizes are inhomogeneous throughout the field of view, being progressively laterally stretched out with increasing depth. The quantification of this behavior prior to any regularization is observable in Fig. 2, demonstrating the increasing lateral speckle size for increasing depth. In this same figure, the effect of regularization is clear through the decrease in depth-dependence of the lateral speckle size (the linear fit through the data is more horizontal after reg- ularization) and the similarity between the axial and lateral speckle sizes (around 1 mm), indicating near isotropy of the speckle dimensions. Important to note is that in the lateral dimension, some (although minimal) depth-dependency still persists after regularization, therefore showing some room for future improvement. B. Concentration-intensity calibration In the analyzed patients, > 95% of the recorded acoustic intensity values are below 1200 a.u., hence calibrating this range takes priority over the higher measured concentrations. From the measured concentrations in the phantom experiment, this range of acoustic intensities is covered by the lowest concentrations up to 1.00 mg/L, which is observable in Fig. 3. Here, the linear fit to the data has an R2 value of 0.94, indicating a good fit quality. This indicates that indeed lower UCA concentrations can be linearly related to their respective measured acoustic intensities. C. CUDI On standard B-mode ultrasound it proves very difficult to determine the extent of adenomyosis in the uterus, specifically identifying the microvascularity and boundaries of the affected tissue. In contrast to this, when observing the same area in CEUS mode, the order of enhancement and the vascularity (which is more tortuous, dense and exhibits angiogenesis in adenomyotic tissue) provide a significantly clearer picture of the extent of adenomyotic tissue [2]. Yet, the observation of the uterus in B-mode and CEUS remains a qualitative assessment. Based on CEUS, CUDI proves to be a pow- erful framework to quantify the UCA dispersion kinetics and translate this to microvascular architecture. Hence, CUDI temporal fitting and spatiotemporal similarity yield quantita- tive structural knowledge on the microvasculature and since this directly corresponds to the formation of adenomyosis, CUDI is capable of both localizing the adenomyosis-invoked angiogenesis and quantifying its extent. As is observable in Fig. 5, a healthy uterus yields more homogeneous and lower values for κ and the spatiotemporal similarity metrics, which translates to higher levels of local dispersion ( κ and the level of similarity are both inversely proportional to the level of Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply. Fig. 1. Local 2D autocovariance (right) for the selected region in the far field of the unregularized CEUS image (left). The speckle size anisotropy can be easily observed in the unregularized autocovariance. Fig. 2. Speckle dimensions before (blue) and after (yellow) regularization, including axial and lateral linear fits. Observe the more isotropic speckle size (comparable axial and lateral dimensions after regularization) and decreased depth-dependence. Fig. 3. Linear fit ( R2 = 0.94) to the measured mean acoustic intensities of the phantoms with the concentrations 0.15; 0.30; 0.50 and 1.00 mg/L. Fig. 4. CUDI temporal fitting ( κ) and spatiotemporal similarity parametric maps of the adenomyotic uterus, after implementation of the proposed speckle size regularization. The myometrium is delineated in white on the B-mode image, with therein the endometrium. local dispersion). In stark contrast to this, an adenomyotic uterus demonstrates higher values for κ and the spatiotem- poral similarity metrics, and more heterogeneity therein, as observable in Fig. 4. This increase in κ and spatiotemporal similarity translates to lower levels of local dispersion and the heterogeneity thereof to the diffuse nature of adenomy- otic anatomy. To quantitatively support these observations, Fig. 6 demonstrates the increased κ, correlation, coherence and mutual information for the adenomyotic myometrium in comparison to the healthy myometrium. Hence, the difference in the CUDI quantitative parameters between adenomyosis and healthy myometrial tissue is evident from both qualitative and quantitative observations of these uteri, but require further analysis in the future on a larger population. Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply. Fig. 5. CUDI temporal fitting ( κ) and spatiotemporal similarity parametric maps of the healthy uterus, after implementation of the proposed speckle size regularization. The myometrium is delineated in white on the B-mode image, with therein the endometrium. Fig. 6. Distributions of κ, correlation, coherence and mutual information for the aforementioned adenomyotic myometrium (Fig. 4) and the healthy myometrium (Fig. 5). Observe the higher values and larger ranges for the adenomyotic tissue in comparison to the healthy tissue. V. C ONCLUSIONS Adenomyosis is difficult to assess using standard B-mode ultrasound, hence 2D CEUS was explored as an alternative to image adenomyosis-related formations of angiogenesis and the general microvascularity. Prior to quantitative assessment of the CEUS acquisitions, the speckle size was regularized to be isotropic and depth-independent throughout the field of view. Also, the relationship between measured acoustic intensities of the backscatter signal and lower UCA concentrations was found to be linear. Then, microvascularity was quantitatively assessed using CUDI temporal fitting and spatiotemporal similarity analysis. In adenomyotic tissue, heterogeneously higher κ and spatiotemporal similarity was observed, indicat- ing decreased local dispersion and an increase in angiogenic development of the microvascularity. Both findings support the hypothesis that the heterogeneity and decreased local dispersion of the adenomyotic vascularity will be reflected in CUDI analysis. ACKNOWLEDGMENT This work has been carried out in collaboration with the Eindhoven University of Technology (Eindhoven, the Nether- lands), the Amsterdam University Medical Center (Amster- dam, the Netherlands), Bracco Suisse SA (Geneva, Switzer- land) and Samsung Medison (Seoul, Republic of Korea). REFERENCES [1] L. Garcia and K. Isaacson, “Adenomyosis: Review of the Literature,” Journal of Minimally Invasive Gynecology, vol. 18, pp. 428—437, 2011. [2] B. Stoelinga, L. Juffermans, A. Dooper, M. De Lange, W. Hehenkamp, T. Van Den Bosch and J. Huirne, “Contrast-Enhanced Ultrasound Imaging of Uterine Disorders: A Systematic Review,” Ultrasonic Imaging, vol. 43, issue 5, pp. 239—252, 2021. [3] M. J. Harmsen, C. F. Wong, V . Mijatovic, A. W. Griffioen, F. Groen- man, W. J. Hehenkamp and J. A. Huirne,“Role of angiogenesis in adenomyosis-associated abnormal uterine bleeding and subfertility: a systematic review,” Human Reproduction Update, vol. 25, pp. 647–671, 2011. [4] M. Mischi, M. P. J. Kuenen and H. Wijkstra, “Angiogenesis imaging by spatiotemporal analysis of ultrasound contrast agent dispersion kinetics,” IEEE transactions on ultrasonics, ferroelectrics, and frequency control, vol. 59, issue 4, pp. 621–629, 2012. [5] M. P. J. Kuenen, T. A. Saidov, H. Wijkstra and M. Mischi, “Contrast- Ultrasound Dispersion Imaging for Prostate Cancer Localization by Improved Spatiotemporal Similarity Analysis,” Ultrasound in Medicine & Biology, vol. 39, issue 9, pp. 1631—1641, 2013. [6] S. G. Schalk, L. Demi, N. Bouhouch, M. P. J. Kuenen, A. W. Postema, J. J. De La Rosette and M. Mischi, “Contrast-enhanced ultrasound angiogenesis imaging by mutual information analysis for prostate cancer localization,” IEEE Transactions on Biomedical Engineering, vol. 64, issue 3, pp. 661—670, 2016. [7] C. K. Mannaerts, M. R. Engelbrecht, A. W. Postema, R. A. Van Kollen- burg, C. M. Hoeks, C. D. Savci-Heijink and H. Wijkstra, “Detection of clinically significant prostate cancer in biopsy-na ¨ıve men: direct comparison of systematic biopsy, multiparametric MRI-and contrast- ultrasound-dispersion imaging-targeted biopsy,” BJU international, vol. 126, issue 4, pp. 481—493, 2020. [8] M. P. J. Kuenen, M. Mischi and H. Wijkstra, “Contrast-Ultrasound Dif- fusion Imaging for Localization of Prostate Cancer,” IEEE Transactions on Medical Imaging, vol. 30, issue 8, pp. 1493–1502, 2011. Authorized licensed use limited to: Universiteit van Amsterdam. Downloaded on June 26,2025 at 11:46:51 UTC from IEEE Xplore. Restrictions apply.