Pulsing and Detection Strategies for Contrast-Enhanced Ultrasound: A Narrative Review

Contrast-enhanced imaging has grown significantly in the past two decades. Technology has evolved from imaging based on linear principles to elaborate pulsing and microbubble-specific detection strategies. This review provides a broad overview of the research published on these topics, emphasizing the progress made, current challenges, and future research considerations. We cover the physical and conceptual underpinnings of imaging based on ultrasound contrast agents, focused on pulsing and detection strategies. The techniques proposed are categorized according to the underlying fundamental physical and signal processing principles. We revisit methods that were previously only of academic interest and may now be clinically feasible with advances in computation and hardware. We discuss unmet challenges and opportunities originating from developments in other sub-fields of ultrasound imaging to enable wider clinical adoption of contrast-enhanced ultrasound.


I. INTRODUCTION
C ONTRAST-ENHANCED ultrasound (CEUS) imaging is an essential tool for diagnostic radiology available for clinical indications in the USA, Europe, and Asia [1], [2]. Previously reported articles detail the evolution of ultrasound contrast agents, their chemical composition, clinical indications, and current status [3], [4], [5], [6], [7], [8], [9], [10]. The physical understanding of the interaction of contrast agents with ultrasound (US) has evolved in the past decade, improving CEUS methods. Recent research on CEUS has focused on nonlinear signal detection. Specifically, contrast-specific imaging techniques such as second harmonic, pulse inversion, amplitude modulation, and harmonic power Doppler are now available in the clinic [1].
A large number of techniques reported are not available commercially because of technological and economic reasons. Given recent technological developments, a renewed focus on previously reported techniques is of interest.
The objective of this narrative review is twofold. First, to provide a holistic view of the research in this area, focusing on current challenges and research considerations to enable technological progress and clinical adoption. The second objective is to lower the barrier of entry for new researchers in CEUS. Specifically, we focus on the conceptual basis and signal processing aspects of imaging based on ultrasound contrast agents (UCA), emphasizing pulsing and detection strategies. Three articles with a similar scope were published nearly two decades ago [11], [12], [13]. However, the field has advanced significantly since then. Detailed reviews have appeared on specialized topics such as super-resolution imaging [14]. A recent in-depth review by Averkiou and colleagues emphasizes clinical and instrumentation aspects of CEUS imaging, and discusses selected techniques that are already in clinical use or on the horizon [1]. The present study focuses on complementary topics and discusses largely a different set of articles. Specifically, we emphasize pulsing

B. SUBHARMONIC IMAGING
Subharmonic imaging offers a high CTR because tissues do not generate appreciable subharmonic signal. The insonation pressure should exceed a threshold to observe a subharmonic response, which depends on acoustic pulse parameters [20], [21], [22], [23], [24] and the physical properties of UCA. Buckling and shell rupture contribute strongly to subharmonic and ultraharmonic generation [25]. Chirps, triangular, rectangular, and dual-frequency pulses were shown to enhance subharmonic emission and reduce its threshold [26], [27], [28]. The threshold also decreases with pulse duration and increases if the excitation pulse envelope is tapered [22], [29], [30]. However, long-duration sinusoidal signals degrade the axial resolution. Coded excitation techniques can address this limitation (Section III-A).
Subharmonic imaging achieves higher penetration depth relative to harmonic imaging (at the same transmit frequency). However, threshold behavior limits sensitivity, especially for deeper locations. Further, subharmonic imaging is limited by the transducer bandwidth. The subharmonic response of UCA has a strong size dependence [20], [31]. Daeichin et al. reported a comparison of subharmonic (SH), nonlinear fundamental (NF), and ultraharmonic (UH) imaging [32]. They combined multipulse techniques (section III-B) -pulse inversion (PI), amplitude modulation (AM), and a combination of PI and AM (PIAM) along with SH, NF, and UH imaging. Subharmonic imaging had the best performance at limited (8 mm) depth than NF and UH imaging. The combination of SH/PI had the highest CTR at 8 mm depth. At a depth of 16 mm, NF showed the best performance. With a bandlimited transducer, UH imaging achieved better performance than SH imaging [33]. Further, UH and SH modes can be combined [34]. Figure 2 compares fundamental, SH, UH, and combined modes for IVUS.
Subharmonic imaging was reported for small and large animal imaging [35], [36]. Goertz et al. used SH imaging at 30 MHz transmit frequency to resolve submillimeter vessels in rabbits [35]. Subharmonic imaging was also performed in canines [37] and in breast cancer patients [36].   [36]. Subharmonic signals have also been used for noninvasive pressure estimation [39], [40], [41], [42]. Advances in monodisperse agents and transducer bandwidth can make SH imaging attractive for superficial depth applications.

C. SUPERHARMONICS/HIGH-FREQUENCY TRANSIENTS
Superharmonic imaging employs higher UCA harmonics (3rd, 4th, and 5th combined) [43]. It improves CTR because nonlinear propagation is relatively weaker for higher harmonics [46], [47]. Higher frequencies also enhance resolution at the cost of lower imaging depth. Low-frequency excitation used in this imaging approach causes larger microbubble expansion, which can cause microbubble fragmentation [43]. Bouakaz et al. used a wideband 96-element phased array for superharmonic imaging. This array was capable of operating at 900 kHz and 2.8 MHz [45]. Developing dual frequency transducers for such applications is an active area of research [46], [47]. Superharmonic imaging uses discrete frequency bands, which can cause ''ghost'' artifacts. van Neer et al. proposed a dual-pulse frequency compounding technique to address this limitation [48]. Two pulses with slightly different center frequencies were fired, and the received echoes compounded. The authors reported suppression of ghost artifacts and an improved resolution. However, the frame rate was reduced two-fold. Danilouchkine et al. divided the transducer's transmit aperture and used one-half to transmit at a higher frequency, and the other-half to transmit at a lower frequency [49]. Studies in flow phantoms and mitral valve leaflets showed resolution enhancement and artifact suppression without compromising the frame rate.

D. ACOUSTIC ANGIOGRAPHY
Acoustic angiography uses the high-frequency broadband response of microbubbles to visualize blood vessels with high spatial resolution [50]. It employs low frequency (1)(2)(3)(4)(5) pulses to generate broadband harmonic echoes from microbubbles (>10 MHz) [50]. This technique can potentially distinguish between healthy and tumor tissue based on microvessel tortuosity. Acoustic angiography was used to create a high-resolution 3D map of the microvasculature [51]. Gessner et al. used a confocal dual frequency wobbler transducer to achieve 20 dB CTR in resolving 150-200 µm vessels [50], [52], [53]. Broadband harmonic echoes exploited in acoustic angiography occur primarily due to microbubble fragmentation, which can limit sensitivity [54]. While acoustic angiography benefits from high CTR, the penetration depth is limited because of reliance on high-frequency UCA signals. The clinical applications of acoustic angiography may be limited to relatively superficial organs. The feasibility of acoustic angiography has been demonstrated in pre-clinical models [51], [55] and patients for imaging vasculature in the wrist, hand, and breast [56]. The reader is directed to [51] for a detailed review of acoustic angiography.

E. DIFFUSION IMAGING
Kuenen et al. proposed Contrast Ultrasound Diffusion Imaging [57]. Diffusion refers to spreading of the UCA in the microvasculature because of a concentration gradient. This method quantifies diffusion by measuring the image gray levels over time. Local diffusion was estimated from the modeled relationship between UCA and gray levels. Diffusion parametric images of four patients were compared with histology data. The area under the receiver operating characteristic curve (AUROC) of 0.91 was observed for prostate cancer localization in four patients [57]. Schalk et al. implemented a 3D version of this technique based on spatio-temporal analysis of 4D Dynamic Contrast-enhanced Ultrasound (DCE-US) data for prostate tumor imaging [58]. Instead of model fitting with a convective diffusion model, a spatiotemporal similarity analysis between the spatial time intensity curves (TICs) was reported to assess local dispersion kinetics [59], [60]. Mutual information as a nonlinear similarity measure between neighboring TICs has also been employed to obtain diffusionrelated parameters [61]. 58 VOLUME 3, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.
Sloun et al. developed a method to enable the independent estimation of dispersion and velocity. The vascular network was modeled as a dynamic linear system, and its impulse response was locally estimated using the indicator-dilution curves. The dispersion coefficient and velocity were then extracted through a model-based parameter estimation approach. In a clinical study on prostate cancer localization with 25 patients, an AUROC of 0.84 was reported [62].

A. PULSING STRATEGIES
CEUS techniques typically use low MI (∼0.1) [63]. Approaches based on microbubble destruction have also been reported for CEUS at higher MI [64] (also see Section III-B.9). Pulsing techniques in CEUS can be divided into two broad categories, those that (i) enhance the nonlinear response from UCA (e.g., chirp pulsing) and (ii) suppress the tissue response to enable UCA-specific detection (e.g., pulse inversion).
Coded excitation has been used widely for improving the SNR of traditional B-Mode imaging. Some coded excitation techniques have been reported to enhance the UCA signal without significantly compromising the axial resolution [21], [33]. The excitation pressure is typically limited for CEUS, necessitating SNR enhancement. Imaging modes such as subharmonic and ultra-harmonic imaging (see Section I) require longer-duration pulses. Additionally, the dynamic range used in CEUS (10-30 dB) [51] is considerably lower than conventional B-Mode imaging (∼50-80 dB). Therefore, sidelobe artifacts that are characteristic of codedexcitation may not appear prominently in CEUS.

1) CHIRP-CODED EXCITATION
Chirp-coded excitation has been reported most widely in CEUS [65], [66], [67]. Linear, quadratic, and logarithmic chirps have been explored. Nonlinear frequency modulated chirps are more sensitive to the frequency shifts due to attenuation, which can lead to artifacts when the chirps are decoded [68]. Linear frequency chirps are robust to such artifacts because of their linear group delay. Techniques such as matched filtering are used for decoding, leading to a short pulse length [67]. Chirp-coded excitation has also been used with harmonic and superharmonic imaging [65], [66], [67].

2) BINARY CODED EXCITATION a: GOLAY CODING
Golay codes have a short correlation length, which is exploited for pulse compression using matched filtering. Pairs of sequences with the same length are employed. Summing of the individual autocorrelation functions of each Golay code pair eliminates sidelobes [69]. Leavens et al. combined Golay-based phase encoding with multipulse techniques (pulse inversion and amplitude modulation) and tested imaging performance in vitro [70]. A 6.5-dB improvement was reported versus conventional multipulse excitation. Shen and Shi employed Golay encoding and 3f0 (three times the frequency of insonation) transmit phasing to enhance the SNR of CEUS [71]. Although SNR was enhanced, range sidelobes were observed. These codes involve sharp changes in the phase of the transmitted waveform, and therefore, are limited by transducer bandwidth. Additionally, this approach requires one transmission each for the code pair's two signals, reducing frame rate and introducing potential motion artifacts.

b: HADAMARD CODING
Gong et al. proposed Hadamard-encoded multipulses (HEM) with fundamental frequency bandpass filtering to improve the CTR and SNR [72]. Multipulses encoded with a secondorder Hadamard matrix were sequentially emitted in each of the two transmission events. Fundamental band-pass filtering was performed on the received echoes to suppress residual tissue-related second-order harmonics. Next, Hadamard decoding was performed on the filtered RF echoes, and the decoded RF signals were aligned and subtracted from each other to obtain the nonlinear fundamental energy. Nearly 20 dB CTR was achieved in vitro. Gong et al. combined fourth-order HEM pulse with multiplane wave imaging to further enhance SNR and CTR [73]. Despite substantial SNR and CTR improvement, HEM can degrade spatial resolution. Fourth-order Hadamard decoding was combined with orthogonal decoding and pulse-inversion to improve CTR by 5.4 dB [74].

B. MULTIPULSE EXCITATION TECHNIQUES
For a detailed discussion on commonly used multipulse techniques, the reader is directed to Averkiou [1] and Crocco [75].

1) PULSE/PHASE INVERSION IMAGING
Pulse/phase inversion imaging was first reported for both the contrast-enhanced B-mode and the Doppler imaging [75]. It employs two mutually phase-inverted pulses sent by the transducer in quick succession, and the resulting echoes are summed [74]. The signal from tissue shows an identical response to the phase-inverted pulses. The signal from UCA demonstrates a different response to the phase-inverted pulses, allowing the detection of perfused regions. Pulse inversion is available on commercial systems and has been reported extensively [76], [77]. Pulse inversion also has been combined with several other techniques such as coded excitation and ultrafast imaging [16], [78], which are described in later sections. Motion artifacts in pulse inversion can be reduced by applying correction methods [79]. simultaneously, with their center frequencies differing by 7-10 fold [80]. A low-frequency (0.5-2 MHz) pulse alters the scattering cross-section of the contrast agent. This pulse is chosen based on the resonant frequency of the UCA. A second high-frequency pulse (3-14 MHz) detects the signals generated in response to low-frequency excitation. This approach offers a high spatial resolution, but requires a dualfrequency transducer. Masoy et al. built an annular array for SURF [80] and achieved a CTR of 15-40 dB. Hansen and Angelsen performed SURF imaging using a dual-band linear array with 1.2 MHz and 10 MHz frequency elements [81], achieving a CTR of 8-14 dB. SURF imaging can potentially be used for applications such as breast, thyroid, carotid, and prostate imaging.

3) RADIAL MODULATION
Bouakaz et al. reported microbubble detection based on radial modulation imaging (RMI) [82]. A low-frequency excitation pulse with low amplitude is used to cause microbubble oscillations. A broadband high-frequency imaging pulse pair is then sent to interact with the microbubbles when they are either in the compressed or in the expanded state. Subsequently, the received signals are subtracted. The tissue response to the high-frequency waves does not change due to the low-frequency modulation pulse; thus, the background signal is canceled. However, the differences in the scattering cross-section of the microbubbles when they are compressed vs. expanded results in a residual microbubble response. The performance of high-frequency RMI using a bandlimited IVUS was evaluated in a flow phantom [83]. A CTR improvement of 6.8-15.1 dB was achieved over standard B-mode imaging.
Classical RMI requires the synchronized imaging pulses to interrogate the microbubbles during the compression and expansion states. Moreover, the influence from modulation pulses affects the propagation velocity of imaging pulses [81], causing delay between two echoes from a fixed target, which reduces tissue suppression. A modified radial modulation approach captured UCA oscillations at ultrafast frame rate [84]. The beat frequency between the modulation pulse and the PRF was exploited for CEUS. The modulations (in spectral domain of slow time) produced by this approach were then demodulated to obtain the CEUS frames. This approach circumvented pulse synchronization challenges. Microbubble flow speeds lower than 0.05 mL/min were detected, achieving a CTR of 16 dB in vitro [84]. Sufficient frequency separation is required between modulation and imaging pulses to avoid artifacts [85]. Thus, the frequency selection for the imaging pulse is constrained to higher frequencies, resulting in lower imaging depth. Jing et. al. [86] used a very low frequency modulation pulse (100 kHz) with imaging pulses (≤5 MHz) to overcome this challenge.

4) DIFFERENCE FREQUENCY EXCITATION
Chen et al. reported a strategy that employs two excitation beams with slightly different frequencies [87]. The nonlinear interaction of the two excitation frequencies with microbubbles produces acoustic emissions at the difference frequency and two times the difference frequency. The authors showed that the harmonic signal of the difference frequency signal is generated primarily by resonant microbubbles.
Frinking et al. reported a method that relies on multiple broadband excitation pulses of high frequency along with another burst to destroy UCA [88]. Imaging pulses were used to visualize the region of interest before and after UCA rupture. Contrast agent was detected by either correlating or subtracting the echoes received from imaging pulses, with up to 22 dB enhancement in vitro.

5) AMPLITUDE/POWER MODULATION
Amplitude/Power modulation is a multipulse technique that is commercially available. Two pulses are transmitted such that one pulse is twice in amplitude as the other [89]. The echo received from the low amplitude pulse is scaled by two and subtracted from the echoes obtained using the high amplitude pulse, which cancels the linear tissue response. However, nonlinearity results in a residual contrast-specific signal from UCA perfused regions. Amplitude modulation has also been combined with Doppler imaging [90], subharmonic, harmonic, and coded-excitation imaging [67], [91]. The main limitations of this technique are reduced frame rates and potential motion artifacts. Li et al. employed amplitude modulation and chirp pulses to increase imaging depth without sacrificing resolution [92]. Proprietary multipulse techniques have also been reported, such as cadence contrast pulse sequencing (Siemens), which combines phase and amplitude modulation [93].

6) CHIRP REVERSAL IMAGING
Chirp reversal imaging employs two mutually time-reversed chirps -an ''up-sweep'' (increasing frequency), and a ''down-sweep'' (decreasing frequency) chirp. The summation of the resulting signals results in tissue suppression. However, the ring-down effect of the microbubble population is different for an up-sweep relative to the down-sweep chirp. Thus, the microbubble response is retained, enhancing CTR. Like other multipulse techniques, chirp reversal is limited by reduced frame rate, potential motion artifacts, and needs specialized hardware. This method is combined with traditional filtering strategies to avoid motion artifacts [94], [95], [96].

7) SECOND HARMONIC INVERSION
Pasovic et al. proposed second harmonic inversion to reduce the tissue-generated harmonics and enhance the CTR [97]. Two pulses with the same frequency and amplitude were used, but having a phase difference of 90 degrees at the fundamental frequency. They achieved up to 20 dB suppression of tissue harmonics over the axial range of 30-130 mm, albeit at the expense of a reduced frame rate. More research needs to be performed to assess this technique's performance in vivo and potential artifacts in images.

8) COUNTER-PROPAGATION IMAGING
Multipulse techniques such as PI, AM, and chirp reversal are not entirely specific to microbubbles because of nonlinear propagation. Nonlinear propagation through microbubble suspensions can produce ''pseudo enhancement'' artifacts in distal regions [98]. These artifacts are pronounced even at acoustic pressures lower than 200 kPa. An example is the ''far wall artifact'' while imaging the carotid artery.
Renaud et al. exploited the interaction of two waves propagating in opposite directions for CEUS [98]. The pulses were fired at the transducer bandwidth's low and high-frequency ends. The lower frequency waves were also transmitted without the high-frequency wave as a reference. Counterpropagation imaging exploits that fact that tissues exhibit linear dependence of the elastic modulus on the applied acoustic pressure. Therefore, the counter-propagating pulses are not modified after they pass over each other. However, UCA suspension shows significant nonlinearity even at low pressures, which changes the waves through nonlinear interactions. Comparing the reference signal (when only lowfrequency wave is transmitted) to the echoes received when both high and low-frequency pulses are transmitted enables CEUS. The counter-propagation technique did not produce pseudo enhancement in the regions distal to the microbubble cloud [99]. However, this technique is suited only for superficial targets such as carotid or femoral arteries because of physical limitations in producing counter-propagating waves in situ.

9) DESTRUCTION-REPERFUSION
Destruction-Reperfusion imaging uses a combination of high MI and low MI pulses for CEUS. Disrupting the microbubbles with a MI pulse and then imaging with low MI can provide quantitative information of blood flow within a selected region of interest [100]. In this approach, microbubbles are continuously injected such that the influx of microbubbles equals the steady-state clearance. The microbubbles in a single plane are then destroyed, and the refill rate is monitored over time. The refill rate curve obtained is used to extract parameters such as the time to peak intensity, blood flow velocity, blood volume, and mean transit time, which contain valuable diagnostic information [101], [102], [103]. Parametric imaging has also been reported by the bolus transit method [104]. However, microbubble clearance and dosage restrictions limit repeated measurements with this approach [105].

IV. FILTERING/PULSE COMPRESSION APPROACHES A. MATCHED AND MISMATCHED FILTERING
Matched and mismatched filtering have been used for pulse compression in medical ultrasound imaging [68]. In matched filtering, an a priori estimate of the transmitted signal is correlated with the received echoes. Most CEUS approaches employ nonlinear modes; thus, the filter is matched to the nonlinear frequency band. Pulse compression in matched filtering is proportional to the time-bandwidth product of the filter. Although side lobes are observed, mismatched filtering by tapering the matched filter reduces this artifact [68]. However, this approach causes main-lobe broadening and reduces axial resolution.

B. FRACTIONAL FOURIER TRANSFORM/FAN CHIRP TRANSFORM
Techniques such as Fractional Fourier Transform and Fan Chirp Transform have been investigated to overcome the limitations of pulse compression by matched filtering [106], [107]. The use of short pulses for superharmonic imaging (wideband excitation) enhances axial resolution. However, a significant spectral overlap is observed in the generated harmonics, which makes filtering challenging. Further, this spectral leakage causes range side lobes after pulse compression. These techniques map signals of interest into new domains that facilitate pulse compression. Fractional Fourier Transform analyses signals by rotating the waveforms to a domain between time and frequency [67]. Fan Chirp Transform (FChT) uses a fan-shaped kernel. It reshapes the time-frequency plane into a fan geometry. Accordingly, the FChT compresses a linear chirp along with its harmonics. Therefore, FChT is particularly useful for second harmonic imaging. FChT can achieve over 50 dB range sidelobe suppression in superharmonic imaging with chirps [108].

C. VOLTERRA FILTER
Volterra filter (VF) is a nonlinear series that incorporates memory [109]. It decomposes the input into the sum of linear and other higher-order nonlinear filter components. The total number of filter components determines the order of the VF. Once the filter parameters of the finite order VF are fine-tuned, they can be used to obtain corresponding linear, quadratic, cubic, and other higher order filter responses.
The filter coefficients for second-order VF were obtained by solving a minimum-norm least-squares (MNLS) problem using equations derived from beamformed RF data [109]. The quadratic component of the VF enhanced CTR without sacrificing spatial resolution. Compared to harmonic images, the quadratic VF images retained the low scattering regions of interest due to their vast dynamic range. Du et al. used an adaptive third-order VF to detect UCA [110]. The images acquired using the quadratic and cubic component provided up to 2 to 7 dB higher mean CTR values than B-mode and PI. The VF can extract nonlinear parts of echo data across the entire spectrum, rather than at the fixed bands, resulting in enhanced resolution. Limited work has been reported using this approach; further studies are required in vivo and in clinical conditions.

D. PHASE-LAG IMAGING
Cumulative Phase Delay (CPD) is observed between the fundamental and second harmonic components when ultrasound passes through UCA, but is absent in tissue [111]. Phase lag VOLUME 3, 2023 was exploited by Demi et al. to improve CTR. The CPD imaging approach outperformed speed of sound-based detection for ultrasound tomography imaging in vivo. CPD imaging offers better temporal resolution than multipulse techniques because it relies on a single transmission. However, bandwidth and UCA size distribution and concentration may affect the phase delay [111], [112].
Under an acoustic pressure of tens of kPa, the transition of phospholipid shell from stiff to buckled state affects UCA scattering. Tremblay-Darveau et al. demonstrated the importance of this mechanism in amplitude modulated contrast pulse sequences [113]. They measured CTR in AM, PI, and AMPI (amplitude modulation pulse inversion) in vitro. Marmottant model simulations showed a negative phase lag in microbubbles undergoing shell softening and a positive phase lag in microbubbles undergoing shell hardening. AM pulses produced 10-15 dB stronger nonlinear signal than PI, for all three microbubble types. At lower insonation pressures (<200 kPa), AM and AMPI provided better CTR than PI. The phase lag caused imperfect cancellation between the half and full amplitude pulse echoes, resulting in a robust nonlinear signal 10-20 dB higher than the second harmonics.

E. TIME REVERSAL ACOUSTICS
The time invariance of the wave equation in a nonlinear regime can be exploited for tissue harmonic cancellation [114]. The time-reversal (TR) technique uses time invariance for perfusion imaging in deeper organs. Couture et al. reported TR combined with PI [115]. These authors used PI with opposite phases correcting for the frequency-dependent phenomenon and then using TR for imaging. Instead of getting the TR signal theoretically [116], they acquired echoes from a copper wire and used them for imaging after time reversal. TR maintained the contrast at high pressures, but injected additional second harmonics at low pressures. The echoes at the onset and end of the pulses remained after applying PI and TR. This approach reduced the tissue harmonic signal by 17.6-32 dB [115].

F. DECORRELATION
In CEUS, the term 'decorrelation' refers to exploiting the loss of correlation in the echo signal from microbubbles and background. Suppressing the echo signal from highly echogenic tissue can be challenging using nonlinear imaging methods alone. Herbst et al. proposed the combination of pulse inversion and acoustic radiation force (ARF) to successfully correlate signals from MBs and tissues. This approach was termed ARF decorrelation-weighted PI (ADW-PI). Pulse inversion images were combined with interframe signal decorrelation data (Figure 3). ARF caused MBs to displace causing echo decorrelation, which was exploited for CTR improvement. Also, the decorrelation data was combined with harmonic filtering methods to obtain enhanced CTR [117]. This approach has also been reported for 3D velocity and flow measurement in blood vessels using CEUS [105].

G. WAVELETS
A Doinikov model-based bubble wavelet transform technique [118] was reported to boost the contrast from the microbubble region relative to the tissue region. A mother wavelet known as a bubble wavelet was formed, based on simulations of the response of a microbubble to ultrasound. Then, a convolution was performed between the bubble wavelet and the signal processed at various scales. The bubble wavelet transform (BWT) yields a sequence of wavelet coefficients that relate the wavelet and the received signal at a given scale. This approach improved contrast because of the similarity of the frequency range of the bubble wavelet and the UCA signal. Further studies are needed to assess the potential of these techniques in the clinical setting.

H. SPATIOTEMPORAL CLUTTER FILTERING
Tissue signals demonstrate high spatiotemporal coherence unlike UCA signals. Therefore, Singular Value Decomposition (SVD) can differentiate UCA from tissue, especially with ultrafast acquisition [120]. Spatiotemporal filtering was also reported with Non-Local Means (NLM) filtering and accurate monitoring of UCA was achieved with bipartite graphbased tracking [121]. This NLM based spatiotemporal filter helped in retaining the microbubble tracks generated by its movement and reduced background noise. Ito et al. proposed a technique based on spatiotemporal analysis of the pixel 62 VOLUME 3, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.
intensity variation over several frames [122]. The variance in intensity differences of consecutively sampled smoothened frames was evaluated for each pixel. Next, masking and thresholding of each variance image frame was performed with respect to a binarized image frame obtained using the discriminant analysis method. The feasibility of visualizing blood vessels in the mouse tail was demonstrated using the proposed approach [122].

I. EMPIRICAL MODE DECOMPOSITION
Empirical Mode Decomposition (EMD) can be used to differentiate nonlinear signals from UCA and tissue [123]. EMD is a fundamental part of the Hilbert-Huang transform (HHT). HHT is suitable for analyzing nonlinear and non-stationary signals and involves two steps: EMD and Hilbert analysis. EMD decomposes the data into several Intrinsic Mode Functions (IMFs). IMF is a term given to the signal's oscillatory mode that represents the signal's highest local frequency. To extract the oscillatory mode, EMD begins by defining local maxima and minima of all signals, which form upper and lower envelopes. The average of the two envelopes is measured and subtracted from the original signal. This subtracted part is the IMF function, and the residue obtained is used to repeat the process, known as sifting. The process is considered complete when the difference in standard deviation between the consecutive observations of the IMF function or the residue falls below a fixed threshold, or when no further IMFs can be produced. Most noise energy gets decomposed into IMFs, leading to a significant improvement in image contrast.
Liao et al. demonstrated the effectiveness of EMD in second-harmonic imaging to improve CTR. Extensions of EMD, called Ensemble Empirical Mode Decomposition (EEMD) and Bi-dimensional Empirical Mode Decomposition (BEMD), were also reported to enhance CTR [124].

V. ULTRAFAST IMAGING APPROACHES
A variety of ultrafast imaging approaches have been reported. A proprietary ultrafast CEUS technique is also commercially available in SuperSonic™MACH™systems (Hologic Supersonic Imagine).

A. ULTRAFAST CEUS DOPPLER
Ultrafast Doppler imaging has been combined with CEUS, for assessment of flow and vessel mapping [125], [126]. Bruce et al. used a plane-wave acquisition approach combined with nonlinear Doppler processing at 15 MHz [127]. This approach increased microvascular blood flow resolution while compensating for the weaker high-frequency microbubble response. An amplitude-modulated sequence was generated for each angle of a multi-angle plane-wave sequence, resulting in a nonlinear-Doppler pulse, which was repeated to create a nonlinear sequence. The pulsing sequences were summed to create a nonlinear image for each angle, which were then compounded. Desailly et al. used SVD processing on linear ultrafast ultrasound images for CEUS Doppler [126]. For each dataset, the singular value thresholds were determined, and they distinguished singular vectors with high spatiotemporal coherence from those with lower coherence. Subsequently, the consecutive frames were summed as for nonlinear Doppler data after being filtered by SVD.
Traditional contrast techniques (e.g., PI) and Doppler with long ensembles can be combined at high frame rates to enhance microbubble detection [78], [128]. Long Doppler ensembles can be used with contrast agent-specific techniques without significantly affecting microbubble scattering coherence at low MI. Repeated insonation can change the acoustic response from microbubbles over time, degrading the Doppler signal and decorrelating the echoes. Using low insonation pressures can reduce this effect. Quantitative Doppler imaging has also been combined with amplitude modulation to visualize microvasculature with up to 17 dB higher CTR than PI-Doppler [90]. VOLUME 3, 2023
A Super-resolution Optical Fluctuation Imaging [142] (SOFI)-inspired approach exploits the movement of microbubbles along the vessels to induce correlated fluctuation in adjacent pixels along the streamlines and uncorrelated variations in different vessels. Consequently, SOFI enables the differentiation of adjacent blood vessels and a smooth depiction of the blood vessels along the streamlines. AI-based approaches inspired by optical microscopy have also been reported [145]. Deep learning (DL) may be used in ULM imaging to boost imaging efficiency and implementation flexibility [147]. By using the sub-pixel convolution layer, the computational and memory complexity can be reduced. The image resolution can also be improved by using normal up-sampling and transposed convolutions, which are implemented using an encoder-decoder network, similar to the U-Net architecture [146]. Liu et al. used a modified subpixel convolutional neural network in ULM (mSPCN-ULM), which increased data processing speed and accuracy [147]. DL is also useful in the spatiotemporal filtering of USR images. A three-dimensional convolutional neural network (3DCNN) was reported to reduce clutter significantly [148]. The main challenges of super-resolution imaging with respect to clinical translation are acquisition times and high data volumes. The reader is directed to [14] and [136] for a more exhaustive review of USR.

VI. FUTURE PERSPECTIVES
Physical insights from modeling and laboratory experiments on UCA have led to new CEUS imaging approaches. Many of the reported approaches have shown technical feasibility, but have not been evaluated in vivo and in human trials. Adequate funding and investment is needed for continued development of this technology in animals and clinical sys-tems. The global CEUS market is currently USD 1.8 billion in 2020 and is expected to grow at a compound annual growth rate of 4.0% from 2021 to 2028 [138]. The USA has the largest ultrasound market, but only a limited number of FDAapproved applications exist for CEUS, which has reduced growth. Off-label use is prevalent but not typically covered for reimbursement [150].
Another challenge is the lack of data on many techniques in conditions representative of clinical imaging. In most studies, work is reported using homogeneous tissue-mimicking phantoms. The attenuation coefficient of the phantom and the depth of the target may not model realistic clinical conditions. Even studies on small animals typically consider an ''optimistic scenario'' because of the tumors' superficial depth, unlike clinical tumors that are usually seated several centimeters deep. Further, the heterogeneity of tissue can degrade images because of attenuation, phase aberration, and reverberation clutter [26].
Deeper tissue necessitates the use of lower frequencies, which reduces imaging resolution. The lower excitation pressures used in non-disruptive techniques reduces SNR [151]. Coded pulsing, pulse compression, and beamforming techniques that are specific to CEUS can be investigated further to address this challenge. Synthetic transmit focusing techniques could be employed to reduce MB disruption, albeit at the cost of potential motion artifacts [152], [153]. Hardware and computational enhancements can enable realtime implementation of methods that were previously considered impractical. For example, we may expect improved transducer bandwidths using CMUT transducer arrays, and real-time 3D CEUS using graphics processing unit (GPU) computing. Lower-end systems and specialized applications, such as IVUS imaging, may continue to rely on techniques that require simple hardware and processing.
Most contrast-specific techniques either need arbitrary waveform generation or perform better with this capability. The excitation circuitry of commercial ultrasound systems rely on either square wave pulsers or linear amplifiers [154]. Linear amplifiers can output arbitrary waveforms without introducing harmonic distortion and are thus advantageous for CEUS [155], [156], [157]. However, most commercial systems use square wave pulsers because they are simple, inexpensive, compact, and power efficient [154], [158]. Advances in microelectronics may make it cost-effective to incorporate pulse amplifiers, enabling a wider range of approaches to be translated.
In the next decade, techniques based on plane wave (ultrafast) imaging may become more widely available on clinical scanners. Applications of super-resolution imaging are still emerging that could enable insights into disease mechanisms, progression, and therapeutic assessment. The narrow field of view of super-resolution imaging may be addressed through advancement in 3D imaging. Theoretical considerations suggest up to ten-fold improvement in resolution over the ultrasonic diffraction limit, however only up to five-fold enhancement is typically achieved.
Advances in microbubble detection, isolation and motion correction algorithms can help improve resolution further. Detection and localization of individual bubbles in microvessels using nonlinear imaging approaches may be an important area of investigation. Further, approaches such as sparse signal processing and beamforming could be used to reduce computational overhead. Application of machine learning in super-resolution imaging will likely grow rapidly in the near future [159], [160]. Data complexity reduction, and employing higher contrast agent concentration will help translate super-resolution imaging to the clinic. Competing approaches such as photoacoustics imaging also exist for imaging vasculature. More research is necessary to establish clinical indications for super-resolution imaging.
In the near future, monodisperse contrast agents and contrast agents designed for specific imaging modes may become available clinically. Approaches for quantitative estimation and system independent assessment of contrast agent perfusion will also be a likely area of advancement. Improvement in the understanding of complex contrast agent behavior such as bubble-bubble interaction and multiple scattering may help improve quantitation of CEUS parameters.
The performance of CEUS also depends on the type of probe used. User-defined settings, such as dynamic range and gain, also affect performance, and thus, techniques for automatic setting of these parameters may be an area of research inquiry. Consensus would be needed over choice of optimal imaging parameters, contrast agent dose, and preferred clinical indications for techniques on the horizon.
Despite the aforementioned challenges, the field of CEUS continues to grow. Previously, many of the reported techniques were considered to be only of academic interest. However, some of these techniques may now find niche applications. Therefore, more preclinical and clinical research is required to translate promising approaches to the clinic. In the next decade, one can expect new technologies becoming available commercially/clinically and enabling the wider application of CEUS in preclinical models, as well as for in vivo diagnosis and treatment monitoring/longitudinal assessment of therapy.