The vessel diameter is depicted accurately when overlapping MBs are treated as single and confirms our hypothesis on the overlapping MB location. These results are in agreement with the synthetic data. By treating overlapping events as a single event, all localizations are found inside the vessel, and the vessel width is depicted accurately. On the other hand, our knowledge on the PSF is not adequate to inform the splitting of overlapping events accurately (similar to Fig. 1H), and therefore they can be located outside the vessel resulting in an overestimation of the vessel width (Fig. 3E). In addition, when only well-isolated single MBs are used, the number of detections drops by 40%, compared with when overlapping events are accounted for (Table 2), which leads to an underestimation of the vessel diameter.
Algorithm Performance Assessment In Vivo
The extracted ultrasound video loops comprised 500 to 1500 (processed) frames with frame rate of 12 to 13 Hz. All videos were acquired in the wash-in or the wash-out period of a bolus injection apart from the videos used for OCT comparison, which were generated by using an infusion of sparse MBs. The mean video duration for bolus injections was 121 ± 19 seconds and for infusions 92 ± 28 seconds.
Figure 4 illustrates an example of the algorithm performance in an in vivo setting depicting the density maps using the original pixel size (132 μm) or the 3 × 3 subdivision of the original size (44 μm) as described previously. Figure 4A shows the tiled microscopy image, lectin stained to show endothelial cells, of a slice of a typical mature CL of the sheep ovary (day 9). The circular pattern is due to the near-spherical shape of the CL. This shape is the result of the rapid angiogenic growth of the CL in the first 9 to 10 days of the oestrous cycle of the ewe. Its vascular architecture here shows that the largest arterial feed surrounds the tissue and branches into smaller arterioles. The size of this artery is between 200 and 300 μm. The arrows in Figure 4A show that the artery is no longer in view in this slice, which is due to continuation and branching off plane. Branching of this artery into arterioles is evident across the entire length of these arteries. The core of the CL is mostly filled with microvessels. Not shown in Figure 4A is the supporting ovarian vessels (artery and vein) that lie under the location of the CL. Figure 4, B and C are the B-mode and peak contrast image in CEUS mode, respectively. The MB movement was tracked through the CL, 27 seconds after the peak contrast image and when the population of MBs was sparse enough for the algorithm to operate. The video loop consists of 583 frames, and its duration was 48.6 seconds. In Figure 4D, the processing includes the inverted Gaussian, and the parameters are adjusted for single MB detection only. The number of detections were 10,906, the average number of detections per frame was 19 (ranging from 1 to 42), and the number of tracks was 1135. Note that, in the absence of ground truth, the number of missed or spurious events cannot be quantified. Figure 4E displays the path density map that uses the inverted Gaussian in the segmentation process and is set to detect both single and overlapping MBs. This resulted in 34,325 detected events, which is on average 59 events per frame (ranging from 7 to 138), and 2951 tracks. The display has a pixel size 44 μm. Figure 4F shows the corresponding mean velocity map. Figure 4G shows the density map for single and overlapping MBs just as in Figure 4E, but uses the gradient image in the segmentation process instead of the inverted Gaussian. Finally, although Figure 4, D to G used the 3 × 3 subdivision process, the respective density map of Figure 4E with the original pixel size is shown in Figure 4H.
There is good resemblance between Figure 4A and Figure 4E, whereas Figures 4D and 3G depict a number of sparse tracks that bear little resemblance to the vascular architecture of the sheep ovary (Fig. 4A). Arrows in Figure 4E show a very similar location of the termination of the feeder arteries (arrows in Fig. 4A). Further, these larger vessels have a large number of paths (in yellow) compared with the inner CL, which agrees with their large comparative size shown in the histology. In addition, a number of smaller vessels are shown to branch inwards from these feeder vessels as shown in the histology slice. The rest of the inner CL area has a small number of paths that may be attributed to low microvessel flow. Under the boxed area of Figure 4E is the ovarian artery and vein, not shown in Figure 4A. These display the largest density of paths in Figure 4E and largest blood velocity (Fig. 4F). These features are not apparent in Figure 4D and Figure 4G that use only single MB processing and gradient image approach in the segmentation, respectively. In addition, using the original pixel size for localization in density maps (Fig. 4H) cannot depict different types of vessels. The feeder vessels that surround the CL (200 to 300 μm) and other microvessels inside the CL do not appear different to the ovarian artery and vein below the CL that are a couple of mm wide.
Density maps were obtained and processed similarly from 6 contrast video loops, which were typical of the range of data sets that were acquired, and the information is displayed in Table 3. From these, 1 was in wash-in and wash-out of the bolus, limiting the process only within the CL, 3 were in the wash-out and 2 were captured during an infusion of sparse MBs. The duration of each video loop varies from 39 to 129 seconds.
Table 3 shows that detecting both single and overlapping MBs using the new methodology with the inverted Gaussian as the input for the segmentation process maximized the number of detected events. As there is no ground truth in these data, a manual observation confirmed that the proportion of missed (undetected MBs) and spurious events (wrong detections) is low and did not affect the final density map. Indeed, within the CL, there is clear vascular network pattern. This table shows that the overlapping and single MBs using the inverted Gaussian (column 7) was always the maximum number of detected events in each data set and thus optimal detection. When the algorithm was adjusted to count the single events (column 6), it was shown that the overlapping events were between 32% and 68% of the total number of detected MB events. Columns 4 and 5 show the particle underestimation effect of using the gradient image as input for the watershed function instead. When both single and overlapping events were used in the detection process, the number of events accounted for were between 17% and 57% less than the ones that were counted using the inverted Gaussian. This was due to the shrinking of these particles' areas resulting from the gradient image calculation. The shrinking caused a lot of MB events to diminish to sizes that were not more than 5 pixels, which is the minimum input size threshold, and thus were eliminated. In addition, the size discrimination between single and overlapping events became more difficult to implement. This is because the algorithm size threshold was implemented after the segmentation process, and the shrinking due to the gradient image resulted in a lot of overlapping MB events to be misclassified as single. The resulting column 5 is likely to include mostly overlapping events, as many single ones were removed by the minimum input size threshold. In addition, when the particle size parameter was adjusted for single events, then the number further decreased, but a large number of overlapping events remained included (column 4). If it is assumed that the total number of true MB events is approximately that of column 7 and that the single events are reasonably approximated in column 6, it is then possible to produce an estimated number of single MB events accounted for in column 4, which is the difference between the column 5 number and the number of overlapping events (difference of columns 7 to column 6). This varies between 0% and 63% of the total number calculated. Negative numbers in brackets signifies that no single events are accounted for, and a number of overlapping events were also missed. These results confirm the synthetic data behavior observed in Figure 2, D and F.
Validation of In Vivo Ultrasound Size Measurements
Density maps from 10 different ovaries from 8 different sheep were compared with either histology (4 ovaries), OPT (6 ovaries), or OCT (2 ovaries) imaging. From these, 8 were bolus injections and 2 were captured during an infusion. Optical coherence tomography provided a live image of the vessels close to the surface of the ovary in situ, and the comparison with the respective density map is displayed in Figure 5. The main vessels are clearly seen in the density map (Fig. 5A) and the OCT image (Fig. 5B). Table 4 provides a summary of the measurements made. For the images shown in Figure 5, the vessels were measured to have diameters of 0.9 ± 0.01 mm from OCT compared with 1.19 ± 0.01 mm on the density map, and the narrow vessel diameter was measured to be 228 ± 23 μm from OCT compared with 236 ± 15 μm. The mean difference in measured sizes between the 2 was approximately 10%, which is not significant. Thus, OCT provided the best agreement between density maps, showing that the density map (Fig. 5A) can provide quantitative and accurate information of vascular structures, including vessel diameter and bending.
Compared with the OCT, the OPT provided information from larger areas of the tissue and in 3D. Compared with standard optical microscopy, OPT provided a 3D reconstruction of the whole ovary. It was then possible to choose a subvolume at a region of choice, which could also have a similar thickness to the ultrasound scan plane, which was approximately 2 mm, for comparison. This is not possible with standard 2-dimensional (2D) microscopy where the slices have a thickness of approximately 5 μm. Their orientation is roughly estimated before slicing, and therefore the comparison with the ultrasound image is difficult to make. Thus, in the example of Figure 4, it is rather fortuitous that Figure 4A appears very similar to Figure 4E. An example of a density map and the corresponding OPT slice are shown in Figure 6. The structure of the ovary from the ultrasound image can be seen in each modality: (a) peak contrast in CEUS mode, (b) density map in super-resolution processing, and (c) OPT. There are structures in the ovary identifiable in both the density map and the OPT slice such as the follicles (F1, F2), the CL, and larger vessels (V). The 2 follicles are well defined by the density map, and the large CL has a dense vascular network. On the smaller scale in the density map, there is a track detected in the first follicle, which is assumed to be part of the outer surface of the follicle. This is also seen in the OPT image (arrowed). This vessel measures 141 ± 54 μm diameter on the density map compared with 83 ± 7 μm on the OPT.
The main comparison performed between density maps and histology slices was on the measurement of the area of the CL. For 3 density maps and histology pairs, the mean CL area measured 48% smaller on histology than that measured on the density map. For all measurements, the mean difference in sizes measured on OPT is 36% smaller than that measured on the density map. Of the measurements made on the density maps, the smallest measurable regions that could be compared with similar regions on OPT were small vessels and follicle walls. A vessel diameter measurement on OPT at 83 ± 7 μm is compared with 141 ± 54 μm on the density map. Follicle wall thickness was measured to be 173 ± 35 μm on OPT compared with 241 ± 10 μm on the density map. It is evident that histology and OPT provide a significant underestimation of all sizes and area measurements compared with the density maps.
Finding comparable vessels in the optical images that were smaller than 100 μm and could be compared with the same vessels in the density maps was challenging. This is because all the different criterion standard techniques did not provide an abundance of such measurements. The OPT staining was inadequate to delineate arterioles, whereas the OCT resolution and sensitivity are not adequate to show the smallest vessels. As mentioned previously, the 2D microscopy with lectin staining provides such vessels but cannot be matched with the ultrasound maps. A number of vessels below 100 μm were observed, however, in the ultrasound maps and are displayed in Table 5 but were not verified. However, the large number of tracks strongly suggests the existence of vessels. The narrowest vessel measured on the density maps was 55 ± 10 μm, and there are several clear vessel paths between 50 and 100 μm.
Example Map of the Prostate
The MB tracking algorithm was applied to a human prostate with cancer. Figure 7, A and B are the B-mode and a contrast image after the peak of the MB density in CEUS mode, respectively. The acquired data set was typical and of low SNR. The video duration was 136 seconds from which the 115 seconds was the processed time that provided a density and velocity map (Fig. 7). The detection parameter combination was optimized to detect both single and overlapping MBs. A total of 1149 frames were processed, where 612,439 events were detected and created 47,536 tracks. The examination time had to be kept as short as possible; hence, an MB bolus injection was used. As a result, there is high MB density per frame, where the average number of detections per frame was 533.
The marked areas by a rectangle shape, an oval shape, and an arrow in Figure 7, C and D show the tumor areas that correspond to the histology. In the histology (Fig. 7E), cancerous areas are displayed with red color, based on the microscopic analysis of cell differentiation (Gleason's pattern) of whole-mount sections according to Montironi et al.76 Before the histopathological analysis, the prostate was fixated in formalin for 24 hours and sectioned in 4-mm slices. Note that the histological slices and ultrasound imaging planes are not parallel, and one imaging plane can intercept multiple slices. The velocity map provides very good correspondence to the histology particularly slices 5, 6, and 7, whereas the density map provides also a reasonable agreement with these slices.
Super-resolution images under realistic patient imaging conditions were achieved, demonstrating the feasibility of clinical 2D ultrasound super-resolution imaging using a standard CEUS mode. The gain in resolution is at least 5-fold, as vessels under 100 μm were detected at transmit frequency of 3 MHz (λ = 514 μm), and the system resolution here is approximately λ (half the pulse length). The smallest verified vessel width was 60 μm (Table 5), and the unverified detection of small arterioles (55 μm) presented nearly an order of magnitude resolution gain. The synthetic data investigation shows that the MB localization uncertainty can achieve 26 μm accuracy. The use of synthetic data enabled the development of the method into an ultrasound one as the errors induced by several parts of the processing were possible to assess and minimize. This was done by investigating detection efficiency, segmentation accuracy, and subsequently MB localization accuracy. The in vivo results are comparable with the literature in terms of resolution improvement. Experiments in thinned skull of rats in a fixed location provided λ/6 resolution at 20-MHz transmit frequency and using ultrafast scanning.31 Elsewhere, λ/4 resolution was achieved using higher transmit frequencies for identification of tumors.33 A clinical scanner has been used in the initial demonstration of super-resolution imaging in vivo, and under 20 μm resolution was achieved, which is over 5 times the improvement to the system resolution.24 This was performed in an optimal setting to minimize aberration, as only a thin slice of tissue was scanned, a flattened mouse ear, and the depth was also limited to 1 cm, and the tissue and probe were static. Our results that resolved structures with at least λ/8.5 accuracy were performed under conditions that are closer to clinical 2D CEUS using standard CEUS mode, low frame rate, radiology-relevant image depth to investigate a volume of tissue. Thus, compared with the above studies, significantly increased aberration was present in the in vivo data here. This results in additional PSF shape distortion with potentially negative consequences in MB detection and segmentation. Given the approximately 2-mm thickness of the 2D ultrasound plane, it seems surprising that vessels with tens of micrometers in diameter can be visualized. This is explained as the tracking algorithm enables the super-resolved MB localization across the third dimension of the slice thickness. The resulting density and velocity maps are in fact a projection of the ultrasound-exposed 3D volume into a 2D image plane. Thus, provided that scattering events are possible to distinguish, vessel structure is preserved projected and so is blood velocity. This is because different vessels may lie at different angles (ie, from vertical to parallel) in relation to the scan plane. This may result in inaccurate depiction of velocity as the angle is not known. However, in principle, 2D super-resolution imaging in vivo is not hampered by the width of the ultrasound scan slice, and several microvessels are possible to depict in density maps. Although it can be argued that in the future the velocity accuracy may be improved using 3D CEUS, the 2D prostate image here (Fig. 7D) strongly suggests that the velocity accuracy is not a problem as the high blood velocity were detected only around the tumor and correlated well with its area. This may be attributed to the high density of tumor neovessels that also have irregular pattern. This ensures that a large number of vessels are parallel to the scan plane. All this suggests that, in a clinical study, different types of velocity maps need to be tested to identify those that represent closest tumor dynamics in 2D (eg, maximum velocity that may be hypothesized to represent parallel vessel velocities). In addition, tissue motion artifacts do not appear in the super-resolution literature and seem to be well compensated here. Note that the animal experimental setup ensured that nonrigid motion is avoided and that the rigid motion due to breathing is kept in plane with the 2D ultrasound image plane. Thus, the well-established rigid registration provided good compensation. Further, the prostate did not require tissue motion compensation as breathing motion does not affect the position of the tissue.
The short video loop time (approximately 2 minutes) here, which provided adequate data for processing, strongly suggests that a clinically relevant examination time is feasible. However, such reduction of data results in additional challenges for the processing compared with other studies. The short video loop time required the use of a large number of MBs per frame. Table 3 shows the number of the detected events (single-overlap column) and the time for each processing, giving an overview of the average number of detections per frame. Comparing with literature, our algorithm detects, for the case of prostate cancer, 533 events on average per frame in 1149 frames, whereas the corresponding values, for example, in Errico et al,31 are approximately 13 events per frame, for a 75,000-frame data set. As mentioned previously, millimeter-sized vessels have orders of magnitude more blood volume, and thus MB concentration, compared with microvessels. At sparse MB infusion concentrations, this, in theory, results in very few single MBs in the microvascular bed, whereas a lot more and several overlapping MBs will appear in the larger vessels. Indeed, it was found here that more than 32% of detected events are attributed to overlapping MBs, which implies that overall these account for more than 50% of the MBs in the image, as shown in the CL study (Table 4). Previous investigations tend to avoid overlapping MB events.24,31,32 In this case, large data sets are required to depict the entire vascular space under investigation,24,25,31,52 which implies that clinical examination times would be significantly increased. The advantage of excluding overlapping events is that no assumptions are needed to include these events, and the localization accuracy is optimized. However, using only the single events the visualization of large vessels in their entirety may not be depicted accurately. The comparison between Figure 4D (single only) and Figure 4E (overlapping MBs included) showed that, within the approximately 49 seconds video loop time, the inclusion of only single MBs in the processing provided maps that do not include larger arterioles and feeder vessels. In other words, the exclusion of overlapping events provides a systematic error in mapping the vascular bed. The inclusion of overlapping MB echoes in the processing provided a more accurate depiction of the vascular structure with better representation of MB path proportion in different-sized vessels. It is suggested that the path density correlates well with volume flow, whereas the exclusion of overlapping MBs results in a blood volume estimation error for larger vessels. This further strongly suggests that the assumption that most overlapping events are likely to be located in the larger vessels is correct.
The challenge of including large MB numbers in the imaging may be best addressed using high frame rate imaging, which can provide MB scatter overlap and deploy tracking using the autocorrelation method.31 As mentioned in the introduction, such frame rates require a plane wave transmission that provides high attenuation and limited penetration. Further in CEUS, this limitation further increases the variability of the MB detection efficiency across the image due to increased S/N variability and MB destruction. Thus, the detected MB density and path density do not correlate well to red blood cell density and volume flow, respectively, which severely limits quantitative super-resolution maps of vascular dynamics. Here it is demonstrated that the less variable field of the focused transmission at low nondestructive acoustic pressures ensures reasonably uniform MB detection, with high penetration depth up to at least 6 cm (Fig. 7A). In addition, and as mentioned previously in Errico et al,31 an average of 13 events per frame were detected, whereas the method presented here processed, in the case of the prostate (Fig. 7), enables the handling of a larger number of detections as more than 700 events were detected in some frames. The tracking, used here, used a combination of the nearest neighbor approach and knowledge on the MB intensity, suggesting that it is not significantly inferior to the autocorrelation method. The tracking is in effect a sparse recovery method for the location of an MB path that has very few MB detections. In addition, there is no evidence in the literature that suggests that a high frame rate improves the statistics of the processing. Given that a bolus injection requires a minimum of a couple of minutes for the first pass to complete in most organs, it is suggested that it is this MB transit time, along with the dimensions of the vascular bed, that determine the MB number that is adequate to map the entire vascular structure. Here, it is suggested that mapping capillaries may be beyond the capability of image-based methods that aim at high resolution. Thus, vessels of tens of micrometers in diameter are the realistic targets of super-resolution ultrasound in clinical radiology, and most of these may be crossed several times in by the MB concentrations used here, thus providing adequate signal for processing. The inclusion of overlapping MBs is thus necessary. An approach that deals with the cumulative signal from all MBs in each pixel rather than particle events and, thus, uses the localization of spatially nonisolated MBs77 may be argued to include single and overlapping events in the processing. However, it is not evident from this processing what are the MB event areas, and thus it is much more difficult to assess detection and localization accuracies.
The super-resolution density maps presented here are the result of a robust detection and segmentation processes. All of the MBs were used in the tracking process including both single MBs of low intensity and overlapping MB events of large size and intensity. The signal enhancement through PPI and Haar-like features and the noise removal discriminating the background from the foreground make this algorithm capable of detecting even the weakest scatter. Christensen et al24 based the detection of single MBs on the cross correlation of each region in the reference frame and the subsequent frames. The groups that use ultrafast imaging detect individual events based on the correlation of each MB with the corresponding one in the next consecutive frame, due the high frame rate.23,26,31,51 These approaches are not effectively different to the one proposed here. Our approach enables the automatic detection and identification of MB events. Other methods apply the detection in B-mode frames after a similar approach to our background and foreground discrimination,32,35 thus dealing with multiple MB numbers. However, in CEUS, the shape of each scatterer in the image is nonregular and needs to be preserved. As a result, an optimal segmentation, using appropriately the watershed function, may be proposed for accurate localization of each event as well as ensure that nearly all MBs are detected.
Initial images of prostate cancer have been presented here from one patient. Although not conclusive, this initial result seems promising. Both density and velocity maps show good correlation with the histological evaluation. The velocity map suggests that tumor areas have redundant anastomotic vessels due to neovascularization. Extensive research aims at detecting and grading cancer by imaging technology to replace the use of invasive systematic biopsies (standard procedure) and reduce the risk of overtreatment and undertreatment. As aggressive (high grade) prostate cancer is correlated with angiogenesis and increased microvascular density,78,79 the proposed method may represent an asset for improved prostate cancer diagnostics and monitoring.
In future work, immunohistology by, for example, CD31 staining should be used to establish a ground truth of the microvascular architecture and improve over the adopted indirect comparison with the histology. A full study is required to assess this information that is otherwise difficult to compare with histological evaluation. The different microscopic techniques provided a criterion standard. However, these are limited, and this has become apparent through the improved near-microscopic resolution performance of ultrasound super-resolution images. Both microscopy histology and OPT images provided significantly reduced size measurements compared with the ultrasound density maps (Table 5; 41% mean difference), which confirms several other studies. It is known that optical imaging is undertaken ex vivo and after further tissue processing. This processing results in tissue shrinking.80 Further, fixation and histological preparation distort the tissues, and some of the variables required for this are difficult to standardize.81,82
The OCT performed in vivo provided the best comparison of vessel sizes with the density map with measurements within 10%. This confirmed that the measurements in the density maps are very accurate and demonstrated that ex vivo criterion standard comparison may be less appropriate for super-resolution ultrasound development. In addition, this OCT validation is superior to that using an in vitro setup with a capillary or capillary network that has been used previously.23,30,32 The in vitro setup provides lower PSF variability across the image compared with in vivo tissue imaging. This is because significant changes in the speed of sound across tissue affect the aberration and augment PSF variability compared with that of a translucent in vitro setup. This may be the reason that in vitro setups have been used previously to validate velocity estimation,32 as the diameter estimation is not a robust validation for the real imaging in vivo. In this context, the vessel diameter and thus system resolution are better validated in vivo. The 2 different-sized vessels (Fig. 5), which were embedded in tissue, had different vessel diameter and curvature, which provided a convincing validation of the vessel diameter accuracy of the ultrasound density maps at multiple positions in the same image. However, the most significant challenge with the use of the OCT was the matching of its plane with that of the ultrasound image. First, the ultrasound image is of low resolution and thus impossible to compare in situ. Second, the OCT is limited in depth (1.5 mm), which imposed an unusual position for the ultrasound transducer and restricted the imaging plane to very close to the surface of the ovary (and perpendicular to the plane in view for the rest of the imaging). However, it was shown that live techniques are more likely to be of use in the development of similar high-resolution ultrasound-based imaging methods.
In conclusion, a new super-resolution tool, which can be used with current clinical 2D CEUS, was presented. The feasibility was demonstrated in vivo for clinical radiology relevant image depths, with tissue motion and for short examination times under 2 minutes. The potential lies in the identification of regions with abnormal vasculature and particularly malignant tumors.
We also acknowledge the support of Sebastian Schaefer, Thorlabs (Lübeck, Germany) for the loan of and help with Optical Coherence Tomography; Harris Morrison of the IGMM, University of Edinburgh, for invaluable support, advice, and help with Optical Projection Tomography; Joan Docherty for facilitating and animal management for the ultrasound work; Nadia Silva for histology and experiment management; and Linda Nicol for the advice, ovary washing, and fixing.
1. Hoskins PR, Martin K, Τhrush A. Diagnostic Ultrasound Physics and Equipment
. 2nd ed. Cambridge, United Kingdom: Cambridge University Press; 2010.
2. Sboros V. Response of contrast agents to ultrasound
. Adv Drug Deliv Rev
3. Sboros V, Tang MX. The assessment of microvascular flow and tissue perfusion using ultrasound
imaging. Proc Inst Mech Eng H
4. Claudon M, Dietrich CF, Choi BI, et al; World Federation for Ultrasound
in Medicine; European Federation of Societies for Ultrasound
. Guidelines and good clinical practice recommendations for Contrast Enhanced Ultrasound
(CEUS) in the liver - update 2012: a WFUMB-EFSUMB initiative in cooperation with representatives of AFSUMB, AIUM, ASUM, FLAUS and ICUS. Ultrasound Med Biol
5. Dai Y, Chen MH, Yin SS, et al. Focal liver lesions: can SonoVue-enhanced ultrasound
be used to differentiate malignant from benign lesions? Invest Radiol
6. Bartolotta TV, Taibbi A, Midiri M, et al. Indeterminate focal liver lesions incidentally discovered at gray-scale US: role of contrast-enhanced sonography. Invest Radiol
7. Ta CN, Eghtedari M, Mattrey RF, et al. 2-tier in-plane motion correction and out-of-plane motion filtering for contrast-enhanced ultrasound
. Invest Radiol
8. Chung YE, Kim KW. Contrast-enhanced ultrasonography: advance and current status in abdominal imaging. Ultrasonography
9. Huang DY, Yusuf GT, Daneshi M, et al. Contrast-enhanced ultrasound
(CEUS) in abdominal intervention. Abdom Radiol (NY)
10. Brabrand K, de Lange C, Emblem KE, et al. Contrast-enhanced ultrasound
identifies reduced overall and regional renal perfusion during global hypoxia in piglets. Invest Radiol
11. Ten Kate GL, van den Oord SC, Sijbrands EJ, et al. Current status and future developments of contrast-enhanced ultrasound
of carotid atherosclerosis. J Vasc Surg
12. Appis AW, Tracy MJ, Feinstein SB. Update on the safety and efficacy of commercial ultrasound
contrast agents in cardiac applications. Echo Res Pract
13. Schinkel AF, Kaspar M, Staub D. Contrast-enhanced ultrasound
: clinical applications in patients with atherosclerosis. Int J Cardiovasc Imaging
14. Schneider M, Anantharam B, Arditi M, et al. BR38, a new ultrasound
blood pool agent. Invest Radiol
15. Gao S, Zhang Y, Wu J, et al. Improvements in cerebral blood flow and recanalization rates with transcranial diagnostic ultrasound
and intravenous microbubbles
after acute cerebral emboli. Invest Radiol
16. Wang S, Wang CY, Unnikrishnan S, et al. Optical verification of microbubble response to acoustic radiation force in large vessels with in vivo results. Invest Radiol
17. Porter TR, Xie F, Lof J, et al. The thrombolytic effect of diagnostic ultrasound
–induced microbubble cavitation in acute carotid thromboembolism. Invest Radiol
18. Wang S, Unnikrishnan S, Herbst EB, et al. Ultrasound
molecular imaging of inflammation in mouse abdominal aorta. Invest Radiol
19. Hyvelin JM, Tardy I, Bettinger T, et al. Ultrasound
molecular imaging of transient acute myocardial ischemia with a clinically translatable P- and E-selectin targeted contrast agent: correlation with the expression of selectins. Invest Radiol
20. Tang MX, Mulvana H, Gauthier T, et al. Quantitative contrast-enhanced ultrasound
imaging: a review of sources of variability. Interface Focus
21. Pitre-Champagnat S, Leguerney I, Bosq J, et al. Dynamic contrast-enhanced ultrasound
parametric maps to evaluate intratumoral vascularization. Invest Radiol
22. Lassau N, Coiffier B, Faivre L, et al. Study of intrapatient variability and reproducibility of quantitative tumor perfusion parameters evaluated with dynamic contrast-enhanced ultrasonography. Invest Radiol
23. Desailly Y, Couture O, Fink M, et al. Sono-activated ultrasound localization
microscopy. Appl Phys Lett
24. Christensen-Jeffries K, Browning RJ, Tang MX, et al. In vivo acoustic super-resolution
and super-resolved velocity mapping using microbubbles
. IEEE Trans Med Imaging
25. Hansen KB, Villagomez-Hoyos CA, Brasen JC, et al. Robust microbubble tracking for super resolution imaging in ultrasound
. IEEE International Ultrasonics Symposium Proceedings; 2016.
26. Desailly Y, Tissier AM, Correas JM, et al. Contrast enhanced ultrasound
by real-time spatiotemporal filtering of ultrafast images. Phys Med Biol
27. Sboros V, Moran CM, Pye SD, et al. The behaviour of individual contrast agent microbubbles
. Ultrasound Med Biol
28. Sboros V, Pye SD, MacDonald CA, et al. Absolute measurement of ultrasonic backscatter from single microbubbles
. Ultrasound Med Biol
29. Sboros V, Pye SD, Anderson TA, et al. Acoustic Rayleigh scattering at individual micron-sized bubbles. Appl Phys Lett
30. Viessmann OM, Eckersley RJ, Christensen-Jeffries K, et al. Acoustic super-resolution
. Phys Med Biol
31. Errico C, Pierre J, Pezet S, et al. Ultrafast ultrasound localization
microscopy for deep super-resolution
vascular imaging. Nature
32. Ackermann D, Schmitz G. Detection
of multiple microbubbles
B-Mode Images. IEEE Trans Ultrason Ferroelectr Freq Control
33. Opacic T, Dencks S, Theek B, et al. Motion model ultrasound localization
microscopy for preclinical and clinical multiparametric tumor characterization. Nat Commun
34. Couture O, Hingot V, Heiles B, et al. Ultrasound localization
microscopy and super-resolution
: a state of the art. IEEE Trans Ultrason Ferroelectr Freq Control
35. Ackermann D, Schmitz G. Reconstruction of flow velocity inside vessels by tracking
with an MCMC data association algorithm. Proc IEEE Int Ultrason Symp (IUS); 2013:627–630.
36. Foiret J, Zhang H, Ilovitsh T, et al. Ultrasound localization
microscopy to image and assess microvasculature in a rat kidney. Sci Rep
37. Sboros V, Moran CM, Pye SD, et al. An in vitro study of a microbubble contrast agent using a clinical ultrasound
imaging system. Phys Med Biol
38. Gabe IT, Gault JH, Ross J, et al. Measurement of instantaneous blood flow velocity and pressure in conscious man with a catheter-tip velocity probe. Circulation
40. Bar-Zion A, Solomon O, Tremblay-Darveau C, et al. Sparsity-based ultrasound super-resolution
hemodynamic imaging. IEEE Trans Ultrason Ferroelectr Freq Control
41. van Sloun RJG, Solomon O, Bruce M, et al. Super-resolution ultrasound localization
microscopy through deep learning. Ithaca, NY: Cornell University; 2018. Available at: https://arxiv.org/abs/1804.07661v2
. Accessed March 23, 2019.
42. Yang L, Qiu Z, Greenaway AH, et al. A new framework for particle detection
in low-SNR fluorescence live-cell images and its application for improved particle tracking
. IEEE Trans Biomed Eng
43. Wilson RS, Yang L, Dun A, et al. Automated single particle detection
for large microscopy datasets. R Soc Open Sci
44. Yang L, Parton R, Ball G, et al. An adaptive non-local means filter for denoising live-cell images and improving particle detection
. J Struct Biol
45. Butler MB, Thomas DH, Pye SD, et al. The acoustic response from individual attached and unattached rigid shelled microbubbles
. Appl Phys Lett
46. Thomas DH, Looney P, Steel R, et al. Acoustic detection
of microbubble resonance. Appl Phys Lett
47. Butler MB, Thomas DH, Silva N, et al. On the acoustic response of microbubbles
in arteriole sized vessels. Appl Phys Lett
48. Thomas DH, Butler M, Anderson T, et al. The “quasi-stable” lipid shelled microbubble in response to consecutive ultrasound
pulses. Appl Phys Lett
. 2012;101:071601. Available at: https://doi.org/10.1063/1.4746258
. Accessed March 26, 2019.
49. Thomas DH, Butler M, Pelekasis N, et al. The acoustic signature of decaying resonant phospholipid microbubbles
. Phys Med Biol
50. Herbst EB, Unnikrishnan S, Wang S, et al. The use of acoustic radiation force decorrelation-weighted pulse inversion for enhanced ultrasound
contrast imaging. Invest Radiol
51. Couture O, Besson B, Montaldo G, et al. Microbubble ultrasound
imaging (MUSLI). 2011 IEEE International Ultrasonics Symposium, 18–21 October 2011
52. O'Reilly MA, Hynynen K. A super-resolution ultrasound
method for brain vascular mapping. Med Phys
53. Sboros V, Averkiou M, Lampaskis M, et al. Imaging of the ovine corpus luteum microcirculation with contrast ultrasound
. Ultrasound Med Biol
54. Pinker K, Moy L, Sutton EJ, et al. Diffusion-weighted imaging with apparent diffusion coefficient mapping for breast cancer detection
as a stand-alone parameter: comparison with dynamic contrast-enhanced and multiparametric magnetic resonance imaging. Invest Radiol
55. Kapetas P, Clauser P, Woitek R, et al. Quantitative multiparametric breast ultrasound
: application of contrast-enhanced ultrasound
and elastography leads to an improved differentiation of benign and malignant lesions. Invest Radiol
56. Wang S, Herbst EB, Mauldin FW Jr., et al. Ultra-low-dose ultrasound
molecular imaging for the detection
of angiogenesis in a mouse murine tumor model: how little can we see? Invest Radiol
57. Lassau N, Bonastre J, Kind M, et al. Validation of dynamic contrast-enhanced ultrasound
in predicting outcomes of antiangiogenic therapy for solid tumors: the French multicenter support for innovative and expensive techniques study. Invest Radiol
58. Hudson JM, Williams R, Karshafian R, et al. Quantifying vascular heterogeneity using microbubble disruption-replenishment kinetics in patients with renal cell cancer. Invest Radiol
59. Smeenge M, Tranquart F, Mannaerts CK, et al. First-in-human ultrasound
molecular imaging with a VEGFR2-specific ultrasound
molecular contrast agent (BR55) in prostate cancer
: a safety and feasibility pilot study. Invest Radiol
60. Huellner MW, Pauli C, Mattei A, et al. Assessment of prostate cancer
with dynamic contrast-enhanced computed tomography using an en bloc approach. Invest Radiol
61. Mischi M, Turco S, Lavini C, et al. Magnetic resonance dispersion imaging for localization
of angiogenesis and cancer growth. Invest Radiol
62. Yang Y, Li X, Wang T, et al. Integrated optical coherence tomography, ultrasound
and photoacoustic imaging for ovarian tissue characterization. Biomed Opt Express
63. Schindelin J, Rueden CT, Hiner MC, et al. The ImageJ ecosystem: an open platform for biomedical image analysis. Mol Reprod Dev
64. Schneider CA, Rasband WS, Eliceiri KW. NIH Image to ImageJ: 25 years of image analysis. Nat Methods
65. Grazul-Bilska AT, Borowicz PP, Reynolds LP, et al. Vascular perfusion with fluorescent labeled lectin to study ovarian functions. Acta Histochem
66. Boujelben A, Watson M, McDougall S, et al. Multimodality imaging and mathematical modelling of drug delivery to glioblastomas. Interface Focus
67. Martin K. Properties, limitations and artefacts of B-Mode images. In: Hoskins PR, Martin K, Thrush A, eds. Diagnostic Ultrasound: Physics and Equipment
. 2nd ed. Cambridge, United Kingdom: Cambridge University Press; 2010:64–74.
69. Diamantis K, Anderson T, Butler MB, et al. Resolving ultrasound
using minimum variance beamforming. IEEE Trans Med Imaging
70. Kaplan D, Ma Q. On the statistical characteristics of log-compressed Rayleigh signals: theoretical formulation and experimental results. J Acoust Soc Am
71. Michailovich OV, Tannenbaum A. Despeckling of medical ultrasound
images. IEEE Trans Ultrason Ferroelectr Freq Control
72. Perperidis A, Thomas D, Averkiou M, et al. Automatic dissociation between microvasculature and larger vessels for ultrasound
contrast imaging. Conf Proc IEEE Eng Med Biol Soc
73. Song P, Trzasko JD, Manduca A, et al. Improved super-resolution ultrasound
microvessel imaging with spatiotemporal nonlocal means filtering and bipartite graph-based microbubble tracking
. IEEE Trans Ultrason Ferroelectr Freq Control
74. Yamanaka M, Smith NI, Fujita K. Introduction to super-resolution
microscopy. Microscopy (Oxf)
75. Chenouard N, Smal I, de Chaumont F, et al. Objective comparison of particle tracking
methods. Nat Methods
76. Montironi R, van der Kwast T, Boccon-Gibod L, et al. Handling and pathology reporting of radical prostatectomy specimens. Eur Urol
77. Bar-Zion A, Tremblay-Darveau C, Solomon O, et al. Fast vascular ultrasound
imaging with enhanced spatial resolution and background rejection. IEEE Trans Med Imaging
78. Brawer MK. Quantitative microvessel density: a staging and prognostic marker for human prostatic carcinoma. Cancer
79. Russo G, Mischi M, Scheepens W, et al. Angiogenesis in prostate cancer
: onset, progression and imaging. BJU Int
80. Kerns MJJ, Darst MA, Olsen TG, et al. Shrinkage of cutaneous specimens: formalin or other factors involved? J Cutan Pathol
81. Leong AS, Leong TY. Newer developments in immunohistology. J Clin Pathol
82. Eisenthal A, Trejo L, Shtabsky A, et al. A novel assessment of the quality of immunohistostaining overcomes the limitations of current methods. Pathol Res Pract
Keywords:Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.
microbubbles; detection; localization; tracking; ultrasound; biomedical imaging; super-resolution; prostate cancer