PAT can perhaps be regarded as the traditional mode of PA imaging as envisaged by early practitioners. It is also the most general and least restrictive PA imaging approach with the fewest limitations on imaging performance imposed by its practical implementation.
In PAT, full field illumination, in which a large diameter pulsed laser beam irradiates the tissue surface, is employed. At NIR wavelengths where tissue is relatively transparent, the light penetrates deeply and is also strongly scattered, resulting in a relatively large tissue volume becoming ‘bathed’ in diffuse light. Absorption of the incident radiation by tissue chromophores leads to impulsive heating of the irradiated tissue volume followed by the rapid generation of broadband ultrasonic waves. These propagate to the tissue surface where they are detected by a mechanically scanned ultrasound receiver or array of receivers. The time-varying detected ultrasound signals can then, with knowledge of the speed of sound, be spatially resolved and back-projected to reconstruct a three-dimensional image. shows three commonly used detection geometries: spherical, cylindrical and planar. Clearly, the cylindrical or spherical detection geometries requires access to all points around the target and are therefore limited to applications such as imaging the breast or small animals such as mice. Planar detection geometries are more versatile providing access to a greater range of anatomical targets, especially those superficially located.
A variety of methods for reconstructing the PAT image from the detected signals have been developed. Conceptually, the amplitude at each point t
in the time record of the PA waveform recorded at a point r
can be regarded as representing the sum of all points in the initial pressure distribution po
that lie on a spherical surface centred on r
and with a radius equal to the product of the sound speed and t
—that is, to say the PA source distribution is regarded as being composed of an ensemble of elemental acoustic point sources each emitting spherical waves. The image reconstruction process can then be thought of as one in which each of the detected PA waveforms are spatially resolved using the sound speed, back-projected over spherical surfaces centred on r
and summed over the image volume— illustrates this for a planar detection geometry. This type of simple ad hoc backprojection is equivalent to delay-and-sum receive focusing or beamforming employed in phased array US imaging. Although it provides a simple and intuitively amenable description of PAT image formation and was used in early implementations [5
], it is non-optimal in terms of accuracy and computational expense.
More advanced methods that provide a more accurate reconstruction and or greater computational efficiency have been developed in recent years, many of which are reviewed in Kuchment & Kunyansky [32
]. These methods can be divided into several categories depending on the type of algorithm employed. Filtered backprojection-type algorithms involve filtering before or after a backprojection step [3
] and can provide an exact reconstruction for spherical [34
] cylindrical [36
] and planar geometries [36
]. Although computationally intensive, they have found practical application for spherical detection geometries used in PAT breast [27
] and small-animal imaging [37
]. A number of series summation-based methods, such as those based on the temporal and spatial spectral decomposition of the detected PA waveforms and a subsequent mapping to spatial frequency components in p0
have been described [38
]. They can provide an exact reconstruction for spherical [40
], cylindrical [41
], planar [42
] and some other geometries such as a cube [39
]. Only in the case of planar or cuboidal geometries can sufficiently fast computational times for practical applications be achieved. In the more practically useful planar case, the implementation involves Fourier transforming the time-dependent pressure data measured over the surface, mapping the temporal frequency to the axial spatial frequency and inverse Fourier transforming to obtain p0
]. It is computationally advantageous because much of the processing is accomplished via the fast Fourier transform and a simple k
-space interpolation via the dispersion relation. This and its ease of computational implementation have led to its widespread practical use [29
]. Time-reversal methods involve computationally re-emitting the measured PA waveforms at each detection point in temporally reversed order by running a numerical acoustic propagation model backwards [46
]. They are perhaps the least restrictive of all algorithms [48
], relying on the fewest assumptions and can be used for any detection geometry, detector distribution and can account for known acoustic heterogeneities. In addition, they can be used to mitigate for signal-to-noise ratio (SNR) and resolution degrading effects of acoustic absorption [49
]. Although memory requirements are modest, for practical use, a fast numerical acoustic propagation model is required since it is necessary to compute the entire wavefield for each temporal backpropagation step. This has perhaps limited its practical application, although a time-reversal scheme using an efficient pseudo-spectral k
-space propagation model has been evaluated using experimental data [49
]. Model-based inversion techniques employ a numerical forward model to simulate the detected PA signals from an initial estimate of p0
or a related quantity [51
]. An improved updated estimate can then be obtained by iteratively adjusting p0
at each spatial point until the difference between the predicted and measured PA signals is minimized. By using matrix inversion methods and pre-computing for a specific geometry as described in Rosenthal et al
], these methods can be fast, albeit at the cost of flexibility and the significant computational expense of the initial pre-computing step.
Inevitably, there are practical limitations. An exact reconstruction usually requires the assumption of an infinite number of wideband omni-directional point-like detectors distributed over a solid angular detection aperture of 4π
sr for a spherical1
or cylindrical detection geometry and 2π
sr for a planar geometry, the latter implying detection over an infinite plane—these aperture conditions mean that the entire acoustic wavefront is recorded, so there is a complete measured dataset. Adequately broadband piezoelectric ultrasound receivers that can provide the necessary megahertz bandwidth are, with some limitations, available. Achieving acoustically small detector element sizes in order to provide a near-omnidirectional response and a spatial sampling interval that fulfils the spatial Nyquist criterion (<λ
/2) at megahertz frequencies is more challenging but possible, depending on the upper frequency limit. The detection aperture requirements can, however, present a more fundamental limitation. For spherical geometries, it is possible, in principle, to completely enclose the source region to fulfil the requirement of a 4π
solid angular aperture. However, for cylindrical and planar geometries, the aperture is always truncated in practice. For planar detection geometries, in particular, measurements are always restricted to a finite region of the infinite plane that the reconstruction algorithm assumes. As a consequence, only part of the wavefront is recorded resulting in image artefacts and reduced spatial resolution—the so-called limited view or partial scan problem [54
]. Artefacts, image distortion and blurring can also arise from sound speed heterogeneities and acoustic attenuation which are not accounted for in most reconstruction methods. Several methods aimed at compensating for sound speed perturbations [48
] and acoustic attenuation [49
] have been demonstrated with the aim of improving image quality in acoustically heterogeneous tissues.
4.1.1. Photoacoustic tomography imaging systems
Spherical scanners: A variety of three-dimensional scanning instruments that employ a spherical detection geometry have been demonstrated. As noted above, for applications such as small-animal or breast imaging that allow the region of interest (ROI) to be enclosed by the detection surface, this geometry offers the highest practically achievable image fidelity on account of the large solid angular detection aperture that can be attained.
shows a spherical scanner design used for PA small-animal [37
] and breast imaging [27
]. The instrument comprises a hemispherical detector bowl with an aperture in the bottom to permit delivery of the excitation laser light. One hundred and twenty eight unfocused 5 MHz 3 mm-diameter piezoelectric elements are distributed in a spiral pattern over the surface. The bowl is mounted on a shaft to allow it to be incrementally rotated with successive excitation pulses so that sufficiently fine spatial sampling can be achieved. A second, smaller-diameter, optically and acoustically transparent bowl into which the breast is suspended is inserted inside the detector bowl. Both are filled with water to provide acoustic coupling. This arrangement allows the detector bowl to be rotated independently without disturbing the breast. The excitation light, sourced from a tunable optical parametric oscillator (OPO) laser system emitting at 800 nm and with a PRF of 10 Hz, is directed up through the aperture in the bottom of the bowl. A near-isotropic spatial resolution of approximately 250 µm over a 6.4 × 6.4 × 5 cm field of view (FOV) was reported. The image acquisition time obtained over this FOV using 240 angular steps per complete revolution was 24 s. shows a maximum intensity projection (MIP) images of the vasculature in the left breast of a patient volunteer with sub-millimetres vessels visible to a depth of 4 cm. The corresponding animated three-dimensional MIP movies available online (see electronic supplementary material, movies S1 and S2) perhaps best illustrate the full extent of the detailed vascular anatomy revealed by these images. The in vivo
penetration depth achieved in this study provides a compelling demonstration of the deep tissue imaging capability of PAT, especially as the optical fluence used was more than one order of magnitude lower than the safe maximum permitted exposure (MPE) for skin.
Figure 4. PAT breast scanner with hemispherical detection geometry . (a) Schematic of system. (b) MIP of the left breast of patient volunteer (lateral projection) over 64 × 50 mm2 FOV. Arrows at top indicate the direction of the incident excitation (more ...)
A spherical detection geometry can also be achieved by arranging the detectors over an arc and then axially rotating the target located within the interior space to synthesize detection over a spherical surface as described in Kruger et al
]. In this study, ex vivo
images of a mouse were obtained. Another small-animal scanner that uses a similar approach has been described in Brecht et al
] and was used to acquire in vivo
whole-body images (). In this scheme, the arc array comprised 64 square (2 × 2 mm) piezocomposite transducers of centre frequency 3.1 MHz distributed over a two-dimensional angular aperture of 152°. When the object is rotated through 360°, this translates to a solid angular detection aperture of 10.6 sr and thus close to the ideal 4π
sr aperture required for an exact reconstruction. The animal is immersed in water, with a diving bell arrangement to allow for the delivery of anaesthetic and respiratory gases and placed at the centre of the array as depicted in . Fibres placed orthogonal to the plane of the array and directed at the mouse provide 755 nm excitation pulses emitted by an Alexandrite laser with an incident surface fluence of 1 mJ cm−2
. To preferentially emphasize the PA waves emitted by anatomical features of different length scales and geometries, the raw RF-detected signals were processed with a family of wavelet and other filters prior to image reconstruction, the latter being achieved using a spherical backprojection algorithm. The spatial resolution was reported to be 0.5 mm and the acquisition time was 8 min based on 150 steps per complete revolution of the animal and averaging over 32 laser pulses. shows a volume-rendered image obtained by the system demonstrating that internal organs, such as the spleen, liver and kidney can be visualised. The benefits of incorporating prior structural information into the formation of the image, in this case via the wavelet filtering referred to above, in order to selectively enhance organs, blood vessels and other anatomical structures are apparent.
Figure 5. PAT whole body small animal scanner based on a spherical detection geometry . (a) Experimental arrangement showing 64 element arc array and fibre delivery bundle. (b) Three-dimensional image of a nude mouse illuminated at 755 nm. Both kidneys are (more ...) Cylindrical scanners
: Although a true two-dimensional cylindrical detection geometry has rarely been implemented for three-dimensional imaging in practice ([63
] being an exception), its one-dimensional equivalent, recording over a circle or an arc to obtain a two-dimensional cross-sectional image, has been widely implemented. Its popularity stems from its ease of implementation and the ability to acquire a high-fidelity image with few artefacts—a consequence of the fact that detection over a full 360° angular aperture can readily be achieved for targets that can be enclosed. If high frame rates are not required, it can also be inexpensively implemented as a laboratory system using a single mechanically scanned receiver and a stepper motor-driven rotational stage. Although widely used, it is, in common with the spherical detection geometry, inevitably limited to applications such as small-animal [64
] or breast [74
shows perhaps the simplest one-dimensional cylindrical scanner for two-dimensional cross-sectional imaging [64
]. It comprises a single 3.5 MHz PZT transducer, focused in the elevation direction to attenuate out-of-plane signals, that is mechanically scanned around the target, in this case the mouse head. The excitation laser light (at 532 nm) is delivered along the axis of rotation in order to transversely irradiate the surface. Although the attenuation of light at 532 nm in tissues is relatively high, as shows, the superficial cortical vasculature can still be visualized with high contrast. The spatial resolution of the system was estimated at approximately 200 µm. Following this early demonstration, a variety of similar single-element mechanically scanned laboratory systems have been used variously to image epileptic events [65
], tumour growth [67
] and cerebrovascular changes [66
] in mice and peripheral joints [76
]. To overcome the long image acquisition times of these systems (minutes to hours), several array-based cylindrical scanners that provide real-time two-dimensional image frame rates have been developed and used to study cerebral haemodynamics [70
], cardiovascular dynamics [69
] and organ perfusion [68
] in mice. One of these systems employs a 512-element array arranged over a full 360° aperture [70
]. With 8 : 1 multiplexing, a single acquisition of all elements could be achieved in 1 s. This permitted dynamic imaging of the wash-in of a systemically introduced contrast agent as it perfused through the superficial cortical vasculature in a mouse. A similar system but employing 64 elements over the arc of a 180° aperture has been used for whole-body small-animal imaging [68
]. Three-dimensional images were obtained by axially translating the target in the z
-direction and concatenating the two-dimensional cross-sectional slices acquired at each axial step. This system was used to acquire structural images of the abdominal, thoracic and heart regions. Its high image frame rate (10 Hz) enabled dynamic events such as motion of the heart chambers and ICG-mediated kidney perfusion to be visualized in real time. Another approach employed a fixed 128-element linear ultrasound array and a rotating target holder to implement a two-dimensional cylindrical detection scheme [77
]. This system was used to obtain ex vivo
PA images of the mouse upper body and both PA and US images of phantoms.
Figure 6. PAT cylindrical scanner for small animal brain imaging  (a) experimental arrangement. (b) Image showing superficial cortical vasculature and a surgically induced lesion. MCA: middle cerebral artery, RH: right cerebral hemisphere, LH: left cerebral (more ...) Planar scanners
: Although spherical and cylindrical detection geometries can provide the large angular aperture required for an accurate image reconstruction, their applicability is constrained by the need for access to all sides of the target. They are not suitable for imaging highly superficial features, such as the skin microvasculature, or if strongly echogenic structures such as bone or lung are situated along the acoustic propagation path. These circumstances call for the more versatile planar detection geometry in which the detection is performed over a finite plane using a two-dimensional ultrasound array or its one-dimensional equivalent, detection over a line using a linear array. PA imaging instruments that employ this geometry begin to resemble conventional diagnostic clinical US scanners, in some cases comprising a hand-held array probe that is acoustically coupled to the skin and moved around while viewing the images in real time. Indeed, a variety of PA imaging instruments use existing commercially available diagnostic scanners, suitably modified so that the RF acquisition can be triggered by the excitation laser in order to detect PA waves as well as US echoes [44
]. Co-registered PA and US images can then be reconstructed either using the hardware beamformer of the scanner or in a post-processing step using a reconstruction algorithm. Thus, the absorption-based contrast provided by the PA image can be exploited to reveal the structure and function of the vasculature while the US image provides information on the surrounding tissue morphology based on its elasto-mechanical properties. In this way, the different but complementary contrast provided by each modality can be used in combination to provide additional diagnostic information.
Although more versatile, particularly for clinical use, image quality provided by a planar detection geometry rarely matches that image quality provided by spherical and cylindrical geometries because of the limited detection aperture. As well as introducing artefacts and distortion, this also reduces lateral spatial resolution—indeed, it is the limited view, not frequency-dependent acoustic attenuation, that tends to limit lateral resolution. Vertical resolution, on the other hand, is relatively independent of the detection aperture and is limited by acoustic attenuation. This results in a disparity between the lateral and vertical resolution and gives rise to an anisotropic spatial point spread function, itself a source of image distortion. Delivering the excitation laser light can also be problematic if an array of receivers rather than a single mechanically scanned receiver is employed. The usual solution is to vertically offset the array from the tissue surface, fill the intervening space with an optically transparent acoustic couplant and deliver the laser light obliquely to the tissue surface beneath the array as illustrated in a. This is readily achievable if a linear array is used for two-dimensional imaging as the excitation laser beams can be delivered orthogonal to the length axis of the array and so only have to ‘clear’ the width of the transducer elements. However, the requirement for a spacer does impose a limitation on the dimensions of a two-dimensional array that can be used—the larger the area, the greater the required spacer thickness which in turn reduces the effective detection aperture and thus image quality.
Figure 7. Use of a conventional ultrasound imaging probe for PA imaging. (a) Two-dimensional array probe. (b) Implementation using a linear array probe for dual mode PAT-US imaging described in Kim et al. . b
shows a specific implementation in which a commercially available diagnostic linear array probe and a pair of fibre bundles are integrated to form a hand-held dual-mode US-PA imaging head [29
]. This system provides co-registered two-dimensional PA and US images at real-time image frame rate (10 fps) and is intended for a relatively deep tissue imaging application: sentinel lymph node detection in the breast. Other similar schemes have been demonstrated for visualizing superficial vascular anatomy, including one that employs a 64-element, 7.5 MHz linear array that also provides real-time laser PRF-limited frame rates (7.5 fps) [44
]. An example of a PA and US image provided by this system is provided in showing the blood vessels at a depth of around 10 mm in the leg. Fronheiser et al
] describe the use of a dual PA–US system that uses a 128-element linear array probe from a commercial ultrasound scanner to image the vasculature in the arm in real time with a view to identifying vessels prior to haemodialysis. A 1.75 D phased array comprising 1280 elements has also been used to provide three-dimensional co-registered PA and US images [80
]. The Visualsonics small-animal ultrasound scanner has also been adapted to provide a dual-mode PA–US imaging capability in an instrument that is now commercially available [81
Figure 8. (left) In vivo PAT images and (right) corresponding US images of a vein at the interior part of the medial lower leg obtained using a linear ultrasound array . The image area is 2.6×2 cm (ticks every centimetre) (a) Cross-sectional image from (more ...)
The use of existing, commercially available ultrasound scanners is a convenient and relatively inexpensive means of implementing PA imaging that exploits the advances in piezoelectric array technology, hardware beamformers and RF acquisition electronics that have taken place in recent years in diagnostic ultrasound imaging. However, PA image quality obtained using these systems tends to be somewhat limited, particularly for superficial imaging applications. This is in part a consequence of the limited detection aperture that linear array-based systems provide. It is also because the transducers used in clinical US scanners tend to operate in the sub-10 MHz range and are resonant to some degree. They are, therefore, often insufficiently broadbanded for the detection of PA waves emitted by superficial structures, the frequency content of which can extend from the low megahertz to several tens of megahertz. To address this a custom designed high-frequency 48-element linear array fabricated from 2–2 piezocomposite elements with a centre frequency of 30 MHz and a 70 per cent fractional bandwidth has been developed for PA imaging [82
]. As with other linear arrays used for PA imaging, it provides a real-time two-dimensional B-scan frame rate, in this case 50 fps. By rapid mechanical scanning, a volumetric frame rate of 1 fps for 166 B-scans was reported and the system used to image the subcutaneous vasculature in the human and rat.
A different approach, based on an optical method of ultrasound detection, has been explored with a view to overcoming the limitations of piezoelectric-based detection for planar geometries. It employs a transparent Fabry–Perot (FP) polymer film etalon that comprises a polymer film spacer sandwiched between a pair of mirrors [84
]. Acoustically induced changes in the optical thickness of the spacer modulate the reflectivity of the etalon which can be detected by measuring the changes in the reflected power of an incident laser beam. By raster scanning a focused laser beam across the surface of the sensor, an incident PA wavefront can therefore be spatially mapped in 2D (a
). There are several advantages of this concept. First, the mirrors of the etalon can be designed to be transparent to the excitation laser wavelength. The sensor head can therefore be placed directly on the surface of the skin, and the laser pulses transmitted through it into the underlying tissue. It thus avoids the detection aperture limitations imposed by an acoustic spacer required for piezoelectric detection methods. It also provides an inherently broadband response from DC to several tens of megahertz and very fine spatial sampling of the incident acoustic field. The effective acoustic element size is defined, to a first approximation, by the diffraction limited dimensions of the focused interrogation laser beam. The notional element size and interelement spacing can therefore be on a scale of tens of micrometres. Perhaps, most importantly, the small element size is achieved with significantly higher detection sensitivity that can be provided by similarly broadbanded piezoelectric receivers of the same element dimensions [85
]. The combination of a transparent sensor head and wideband acoustic performance attributes makes this type of sensor particularly well suited to imaging superficial features located within a few millimetres of the tissue surface. In these circumstances, the short acoustic propagation distances involved mean that the PA signal is only weakly bandlimited by acoustic attenuation and can therefore be extremely broadbanded with a frequency content extending to several tens of megahertz. The requirement for a broadband omnidirectional point detector for an accurate image reconstruction then becomes particularly challenging using conventional piezoelectric detection as it demands element dimensions of a few tens of micrometres. Fabricating piezoelectric elements of such small size with adequate detection sensitivity is problematic since their sensitivity scales with active area. By contrast, the FP sensor sensitivity is, to a first approximation, independent of element size.
Figure 9. (a) Schematic of FP-based photoacoustic scanner used to acquire a three-dimensional image of the vasculature in the palm . (b) Lateral MIP image (top) and vertical (x–z) slice image (bottom) taken along horizontal yellow dotted line on MIP. (more ...)
A schematic of the system is shown in a
along with several representations of the three-dimensional image dataset obtained by scanning the human palm [45
]. The sensor used for this demonstration had a peak noise-equivalent-pressure (NEP) of 0.2 kPa (over a measurement bandwidth of 20 MHz) and a −3 dB detection bandwidth from 100 kHz to 22 MHz. The interrogation laser beam spot size was 64 µm, spatial sampling interval was 250 µm and an FOV was 20 mm. No signal averaging was used. The excitation wavelength was 670 nm and the incident fluence incident on the skin surface was 10 mJ cm−2
, and thus below the MPE of 20 mJ cm−2
for skin at this wavelength [86
shows a lateral MIP and a single vertical slice through the centre of the lateral MIP as indicated. This vertical slice shows the contour of the skin surface as well as several underlying sub-dermal blood vessels. c
shows a volume-rendered representation of the reconstructed image and a series of lateral slices at different depths. These images show the subcutaneous vasculature to a depth of approximately 4 mm—the deepest lying vessel is indicated by the arrow ‘A’ on both the volume-rendered image and the deepest lateral slice. The time taken to acquire the three-dimensional image data shown in was approximately 10 min, limited by the sequential nature of the detection and the low PRF of the excitation laser (10 Hz). There is, however, scope to increase acquisition speed, potentially obtaining real-time three-dimensional image acquisition rates, through the use of higher repetition rate laser systems or parallelizing the sensor read-out scheme using full-field illumination and a photodetector array as described in Lamont & Beard [87
]. As well as imaging the skin vasculature, the system has been used for imaging tumour vasculature [45
], the mouse brain [89
] and embryo [50