The generalized scattered signal from a scatterer O ( x ( t ) ) can be modeled by a polynomial expansion: O ( x ( t ) ) = ∑ m = 1 ∞ a m x m , (4) where x ( t ) is the transmit waveform. The contributions of the nonlinear components are defined by the coefficients a m , whereby for linear systems only a 1 is nonzero. As ultrasound pulses consist of sinusoidal transmit sequences, e.g., x ( t ) = cos ( ω t ) with ω = 2 π f the angular transmit frequency, the nonlinear echo can be approximated by O ( x ( t ) ) ≈ a 1 ⁡ cos ( ω t ) + a 2 2 [ 1 + cos ( 2 ω t ) ] + a 3 4 [ cos ( ω t ) + cos ( 3 ω t ) ] + a 4 8 [ 3 + 4 ⁡ cos ( 2 ω t ) + cos ( 4 ω t ) ] + … (5)

Note from the above equation that even-order terms create echoes at even harmonics (and DC), while the odd-order terms account for echoes at the fundamental frequency and odd-order harmonics. The pulse inversion multi-pulse sequence consists of sending in two transmit pulses that are 180° out of phase with each other (Figure 2A). Upon summation of the resulting echoes s ( t ) , the linear contributions are removed and only even order harmonic signal is retained: s ( t ) =   O 1 ( x ( t ) ) + O 2 ( − x ( t ) ) = 2 ∑ m = 1 ∞ a 2 m x 2 m . (6)

While this technique suppresses fundamental signal, it still requires careful selection of transmit frequency to be able to sensitively detect even order harmonics with the given transducer.

In a similar attempt to preserve nonlinear contributions, amplitude modulation consists of transmitting a sequence of pulses that are scaled by a constant factor. Typically, the echoes received from x 1 ( t ) and x 2 ( t ) = 1 2 x 1 ( t ) (referred to as “full amplitude” and “half-amplitude” pulses respectively) are scaled and subtracted, resulting in a residual signal s ( t ) defined as: s ( t ) =   O 1 ( x 1 ( t ) ) − 2 O 2 ( 1 2 x 1 ( t ) ) . (7)

This results in a signal that partially retains all harmonics, including signal at the fundamental frequency; shown here to third order: s ( t ) ≈ a 2 4 [ 1 + cos ( 2 ω t ) ] + 3 a 3 16 [ cos ( ω t ) + cos ( 3 ω t ) ] + … (8)

It is important to note here that the signal component within Eq. 8 at the driving frequency ω represents the scaled difference in the fundamental component due to different amounts of nonlinear signal in the two driving pulses. This “nonlinear fundamental” signal results from the fact that microbubbles exhibit nonlinear resonance characteristics, specifically an amplitude dependent resonance frequency (Figure 1D). As such, the fundamental microbubble response will not necessarily be linearly proportional to the input transmit pressure, e.g., the response from x ( t ) will not be twice that of 1 2 x ( t ) . Indeed, bubble-specific strategies are currently under development that exploit the accompanying echo phase lag associated with this phenomenon [55]. While this approach retains less even-order harmonic energy than PI, the residual “nonlinear fundamental” is particularly useful as it can be well detected within the transducer bandwidth.

Both PI and AM methods can be performed using three or more pulses, offering some advantages in tissue rejection at the cost of temporal resolution. The combination of these two approaches (PIAM, or CPS) retains similar levels of odd-order nonlinear energy as AM while preserving more even-order harmonics, albeit less than the PI technique alone.

Microbubbles are currently the only clinically approved ultrasound contrast agent. One of the strengths of these bubbles is that they remain intravascular due to their size, allowing for diagnostic measurements that would be otherwise difficult with diffusible tracers. However, there is a growing focus to extend the use of these ‘traditional’ ultrasound contrast agents towards other applications, including molecular-based imaging, imaging of the extravascular space, and as a dual imaging and therapeutic delivery platform.

As microbubble vibrations possess a rich resonant structure (Figure 1B), there have been recent developments towards generating contrast images using microbubble super-harmonic frequency components, defined as third-order harmonics and higher ( n f ;   n = 3,5,6 … ) . An extension of traditional second harmonic imaging techniques, the selective reception of these higher-frequency signals results in higher image resolution and contrast-to-tissue ratios compared to standard contrast imaging sequences. Due to the bandwidth of standard clinical transducers, which limits its ability to transmit and receive signals at both the fundamental and super-harmonic energy bands, the implementation of this approach requires multiple, independent transducer elements. This can be accomplished by designing novel phased arrays with interleaved elements for transmit and receive [109, 110], and confocally aligned dual-element transducers [111, 112]. Recent incarnations of this approach, termed acoustic angiography [113], performs super-harmonic imaging using transmit frequencies between 2–4 MHz and receives echo signal from 25–30 MHz. Using this device, an in-vivo resolution of 150–200 µm and a contrast-to-tissue ratio of 20 dB has been demonstrated [114, 115]. To date, this technology has been employed to image and assess tumor microcirculation [116, 117] and remains mostly pre-clinical; although very recent work highlights its potential for clinical translation [118, 119] and is currently an active area of research.

Local blood pressure estimation provides valuable clinical information on the physiology of many organs, and can be employed in the diagnosis of disease in the heart and kidneys. Most current clinical techniques to assess blood pressure within non-limb vessels use catheter-based manometers, which is an invasive approach and introduces changes to the local blood circulation and thus the blood pressure. Perhaps one of the most impactful applications of non-invasive pressure estimation would be for the early detection of clinically significant portal vein hypertension, defined as an increase in the pressure gradient between the portal vein and hepatic veins exceeding 10 mmHg [120]. As noted almost four decades ago [121], bubble response is a direct function of the ambient hydrostatic pressure and may, in principle, be used as a pressure sensor to detect fluctuations in local blood pressure. An increase in ambient pressure effectively compresses the microbubble, resulting in a shift upwards in resonance frequency. For a given transmit frequency, this will manifest itself in the amplitude of the resulting scattered echo. These original works performed on unshelled bubbles resulted in large uncertainties (as much as 30%, or 50 mmHg compared to reference standards [122]) due to the challenge of detecting the relatively small shift in resonance frequency (∼1 kHz shift from a change in 10 mmHg). While the rheological characteristics of phospholipid encapsulated microbubbles results in much larger resonant shifts (∼0.07–0.24 MHz per 10 mmHg [123]) that may be sufficiently detectable for clinical utility, major advances in this application of remote blood pressure estimation are derived from investigations into the modulation of subharmonic scattering. Based on earlier works on commercially available contrast microbubbles that indicate a decrease in subharmonic scattering with increasing hydrostatic pressure [124], subharmonic-aided pressure estimation efforts (referred to as SHAPE [125]) have met initial success in pre-clinical models [126, 127] and in clinical trials for portal hypertension [128] and intra-cardiac measurements [122].

A flourishing research area within diagnostic ultrasound is the development, implementation and interpretation of ultrafast ultrasound imaging, in which up to 20 kHz frame rates (compared to 10–100 Hz using conventional scanners) can be achieved through advances in hardware and software. This concept is based off the transmission of an ultrasonic plane wave (i.e., unfocused beam), which avoids the time-consuming process of sequential scanning and beamforming conducted by traditional focused-mode imaging. The echoes from a single plane wave transmission are received by the transducer elements and subsequently processed and beamformed in parallel. While the use of a single, unfocused transmit beam results in poor image resolution, SNR can be markedly increased by transmitting multiple plane waves at different angles and compounding the coherent beamformed images. Despite this slight subsequent reduction in frame rate, this still results in a very fast acquisition relative to conventional focused beam, limited in principle only by the two-way speed of sound in tissue. Ultrafast plane wave imaging has opened an array of contrast and non-contrast ultrasound applications that take advantage of such increased temporal resolution, including ultrafast elastography [129], cardiac [130], and Doppler-based applications [131].

Perhaps the most disruptive technique derived from a microbubble-based application of this technology to date is ultrasound localization microscopy (ULM, see Figure 4) [132]. As a super-resolution imaging technique, it has begun a paradigm shift in biomedical ultrasound imaging applications despite many previous investigations into methods to improve ultrasound imaging resolution. In standard imaging techniques, image resolution is bound by diffraction to the scale of the wavelength; for example, in a 6-MHz ultrasound imaging system ( λ = 250  µm), the diffraction limit is 125 µm ( λ / 2 ) . The ULM approach exploits the localization of microbubbles to finely sample and image the microcirculation beyond the limit imposed by diffraction, showing impressive results in the areas of oncology [116, 133] and neurology [134, 135] that result in an improvement of the resolving power of ultrasound up to a factor of 10 compared to the diffraction limit [136, 137]. It is an approach inspired by the light microscopy counterpart; photoactivated localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM). These cutting-edge light microscopy techniques, which can image beyond the diffraction limit by an order of magnitude [138–140], rely on photoactivatable fluorescence probes that display unique spectral features upon exposure to different wavelengths of light. These reversible, “photo-switchable” probes in combination with fast-frame imaging cameras enable the rapid acquisition of frames in which only a subset of the sources is visible. With knowledge of the point-spread function of the imaging system, the collection of many sub-wavelength localizations can be reconstructed with resolution lower than the diffraction limit. Indeed, the development of these techniques was so important that it led to the attribution of the 2014 Nobel prize in Chemistry to Eric Betzig, Stefan Hell and William E. Moerner.

An ultrasonic version of super-resolution is achieved by replacing the fluorescent markers with microbubbles (which are sub-wavelength, individual acoustic sources), and the fast cameras with plane-wave, programmable ultrasound imaging systems. These programmable systems give access to the pre-beamformed time-domain data (RF data), whereby assuming a single source, the signal time delay τ as a function of array position x produced by a single microbubble echo propagating at a constant speed c is given by: τ = z 0 2 + ( x − x 0 ) 2 c   , (10) where z 0 and x 0 are the depth and lateral position of the microbubble, respectively. One approach to microbubble localization is to fit this delay function (i.e., a parabolic function), the peak of which will provide the position of the microbubble at much higher resolution than the wavelength [132]. Alternatively, even on beamformed images acquired from conventional ultrasound scanners, various algorithms have been developed to estimate the intensity-weighted centroid of an individual microbubble and has shown success in dilute microbubble applications [141, 142].

The general concept of acquiring a super-resolution imaging using ULM will next be outlined here. After injection of a dilute suspension of contrast agent, video acquisition of the location of interest, either using B-mode or contrast-specific sequences, can be taken using either conventional beam or fast-frame plane wave techniques. Since the resulting ULM image is constructed point by point, a sufficient quantity of microbubbles is required to reconstruct the vasculature, on the order of 1 million events [135] depending on the vessel density and field of view. Given the relatively slow blood velocities in the microvasculature, this often requires long image acquisition times and results in a vast amount of data for processing. Motion correction algorithms are next applied to minimize motion-related localization artefacts, which present a particular challenge due to these long scan times. Various techniques have been demonstrated within the context of the ULM workflow, including phase-correlation approaches between successive B-mode images, all of which result in corrections on the order of hundreds of micrometers for in-plane motion [143–145]. While out-of-plane motion correction is not possible using this 2D approach, 3D ULM techniques are currently being assessed [146]. Following this, a microbubble-filtering processing step is introduced, which can include isolating nonlinear emissions [134, 141] as well as alternative image processing strategies including spatiotemporal-based filtering algorithms [135, 145, 147]. Microbubble localization is then performed by estimation of its centroid using either the raw RF data or the beamformed image. A critical challenge here is the reliable separation of one microbubble from another. The most direct way of localizing a single microbubble is to use a low concentration of contrast agent (e.g., 10 6  bubbles/ml) [134, 141, 148], which guarantees an inter-bubble spacing (e.g., 100  μ m) of several imaging wavelengths at traditional transmit frequencies. Even in such instances, the robust SNR generated from an individual microbubble is of paramount importance, and will ultimately affect the ULM resolution. Recent work [149] has suggested that exploiting the phase response of vibrating microbubbles, a property linked to their resonant nature [75], can increase ULM image quality. However, there are emerging alternative strategies that allow for higher local doses of microbubbles, attempting to circumvent the spatial resolution versus acquisition time trade-off inherent to ULM. Increased local microbubble concentrations not only shorten the scan time, but increase the SNR. In order to overcome the overlapping of the point-spread functions, spatiotemporal filtering algorithms to separate overlapping microbubble signals [150, 151] have been introduced. Recently, algorithms based on deep learning (Deep-ULM) have been proposed, offering the advantage of acquiring high resolution images with high microbubble concentrations and lower computation load compared to other techniques. This AI-based approach is capable of learning the nonlinear image domain implications of overlapping point-spread functions originating from populations of closely spaced microbubbles [152]. Finally, tracking of microbubble trajectories, using simple or more complex algorithms [145, 153], allows not only for the estimation of super-resolved blood flow velocities [135, 144], but for improved image quality due to the fact that a single microbubble can reconstruct several pixels during its trajectory. Indeed, as adequate sampling of microbubble location is critical for the success of tracking algorithms, ultrafast imaging techniques offer a major advantage over conventional imaging approaches. Images are often then reconstructed by projecting the detected tracks on a sub-wavelength grid matrix. True estimates of vessel diameter, therefore, cannot rely on sparse tracks but require them in sufficient number to ensure mapping of the entire lumen, a track density determined by the width of the vessel divided by the super-resolved pixel size [154].

While still in its infancy, ULM has already provided a new in-vivo approach to the study of tissue pathology, providing quantitative information on the density, tortuosity, and small modulations of flow patterns within the microvasculature at depth. The first clinical applications of this technology, using conventional focused beam acquisition, have been conducted on breast cancer [155], lower limb assessment [156] and liver imaging [157]. While there are still limitations to this approach, including slow scan times, SNR, the use of plane-wave scanners not typical in clinics, large amounts of data storage and processing, and motion artefacts, significant advancements in all of these areas are currently ongoing.

It has long been recognized that ultrasound interactions with biological tissue induce bio-effects of both thermal and mechanical origin [158]. On clinical diagnostic scanners, exposure levels are limited in order to avoid these effects [159]. From a therapeutic standpoint, ultrasound-mediated bioeffects have been investigated as a desired endpoint: with effects ranging from tissue ablation [160], microvascular permeability [161], immunomodulation [162], and vascular occlusion [163]. Recent works have highlighted that microbubble contrast agents, under specific acoustic conditions, can generate a wide spectrum of bioeffects [164–166] that contribute towards the treatment of many diseases. Due to their intravascular nature, a primary avenue of research in microbubble-mediated bioeffects is based on the spatially targeted and temporary enhancement of microvascular permeability, employed to promote local drug delivery to regions of disease. One such promising application is the local and transient opening of the blood-brain-barrier [167, 168] and blood-spinal cord barrier [169, 170] for targeted therapeutics into the central nervous system. This technology has recently entered clinical trials in patients with brain tumors [171–173], Alzheimer’s disease [174] and amyotrophic lateral sclerosis (ALS) [175].

Despite being met with initial success, widespread clinical adoption of microbubble-based therapeutics will require the continued development of online, real-time imaging strategies to guide and control treatments. While some of these applications employ MRI guidance, there is increasing interest in employing the acoustic scattering from the microbubbles themselves as an indicator of treatment outcome. Since the spectral echo characteristics can be indicative of the underlying microbubble vibrations [176], remote detection of these signals during treatment is under investigation as a robust and sensitive tool for therapy guidance. Many preclinical applications of targeted microbubble therapeutics, including cardiovascular disease [177, 178] and cancer [166], are performed as a dual imaging and therapeutic technique. Contrast enhanced ultrasound is applied and interleaved with a therapeutic pulse from either a separate ultrasound transducer [166] or incorporated by way of clinical [179] or custom-designed sequence. In this way, the presence of microbubbles within the anatomical site of interest can be visually confirmed before, during and after the treatment sequences. The acoustic emissions detected during microbubble-based therapies have been identified as potential markers for treatment outcome in applications including blood-brain barrier disruption [180, 181], and targeted therapeutic delivery [182]. To this end, passive cavitation detectors are typically employed to measure raw acoustic data to extract quantitative metrics. Most of these methods to date utilize a single element passive transducer, which does not allow the bubble signal to be localized in space. Ongoing novel engineering of array transducers, combined with passive beamforming algorithms, are currently being designed to spatially map bubble activity and allow for confirmation that elicited bioeffects are localized to the target site [183, 184], see Figure 5. Above and beyond these correlative measures, efforts are underway to establish control feedback algorithms based on the measured bubble acoustic activity to promote safe levels of vibration and avoid more violent, disruptive bubble behaviour that leads to unwanted damage. These algorithms modulate the acoustic transmit parameters based off the real-time feedback from nonlinear microbubble emissions, including sub-harmonic energy [185, 186], harmonic energy [187, 188], or both [189].