Real-time spectroscopic photoacoustic/ultrasound (PAUS) scanning with simultaneous fluence compensation and motion correction for quantitative molecular imaging

For over two decades photoacoustic (PA) imaging has been tested clinically, but successful human trials have been minimal. To enable quantitative clinical spectroscopy, the fundamental issues of wavelength-dependent fluence variations and inter-wavelength motion must be overcome. Here we propose a new real-time, spectroscopic photoacoustic/ultrasound (PAUS) imaging approach using a compact, 1-kHz rate wavelength-tunable laser. Instead of illuminating tissue over a large area, the fiber-optic delivery system surrounding an US array sequentially scans a narrow laser beam, with partial PA image reconstruction for each laser pulse. The final image is then formed by coherently summing partial images at a 50-Hz video rate. This scheme enables (i) automatic laser-fluence compensation in spectroscopic PA imaging and (ii) inter-wavelength motion correction using US speckle tracking, which have never been shown before in real-time systems. The 50-Hz video rate PAUS system is demonstrated in vivo using a murine model of drug delivery monitoring.


Main
Nearly every object has a unique signature based on the optical properties of its molecular constituents. Consequently, optical spectroscopy is one of the most important analytic tools in all of science and technology. For biomedical applications, it can help quantify molecular components within complex solutions and structures based on each constituent's optical absorption spectra [1]. However, optical spectroscopy is not used routinely in vivo because high tissue scattering typically limits coherent penetration to a millimeter or less.
To overcome this barrier, photoacoustics (PA) has been proposed. From Alexander Graham Bell's discovery in 1881 [2], optical absorption has been measured acoustically. PA methods were confined to laboratory devices until the 1990s when several groups leveraged modern pulsed laser technologies to image optical absorbers in vivo deep within highly scattering tissue [3][4][5][6].
The principle is simple [7,8]. Pulsed laser radiation diffusing within tissue is absorbed by different structures, producing local heating proportional to the absorption coefficient. Through thermal expansion, heat generates an ultrasound transient (PA signal) propagating long distances.
Signals are recorded at the tissue surface with an array transducer to reconstruct the absorber distribution. Because signal amplitude is proportional to the light absorption coefficient, molecular imaging within tissue is an important potential feature.
Even with PA's remarkable success, validated clinical protocols have been limited. Indeed, most studies have used small animal models, and associated methods have generally not translated well to humans. Although penetration is limited even for low scattering organs such as the breast, there are many potential applications. Here we address what we believe are the primary challenges limiting its clinical acceptance. imaging: pulsed laser radiation irradiates the tissue simultaneously from all fibers surrounding the US detector, thus creating broad beam illumination. b, Spectroscopic PAUS requires sequential multiwavelength illumination of moving tissue. As shown herein, tissue motion during spectroscopic acquisition can change the concentration of different chromophores (blood and GNR, for example) in a measurement pixel, thus corrupting the evaluated spectrum. c, Wavelength-and depth-dependent optical fluence in tissue can strongly affect the evaluation of optical absorption spectra. In this example, the GNR spectrum changes with increasing image depth. d, Characteristics of state-of-the-art commercial PAUS systems. Top: Pulsed laser pulse repetition frequency (PRF). Bottom left: frame rate using single wavelength, and multiple wavelength with wavelength diversity. Bottom right: applications, direct optical fluence compensation, and inter-wavelength motion correction. Systems include: Acuity (iThera Medical, Germany); Vevo ® LAZR-X (FUJIFILM Visualsonics, USA); LOUISA-3D (TomoWave Laboratories, USA); Imagio (Seno Medical, USA); PAM-03 (Kyoto University and Canon Medical, Japan; System developed by POSTECH and Alpinion Medical, Korea. First, the PA signal amplitude is proportional not only to absorption, but also to laser fluence (i.e., light level at a target) [7,8]. Because tissue attenuation depends on wavelength [33], the absorption spectrum estimated from a PA image can be very inaccurate (i.e., the shape can change dramatically and the wavelength of maximum absorption shift), especially deep within tissue [34,35]. This is illustrated in Figs.1a,c and clearly demonstrated in Results.
To reconstruct the true absorption coefficient distribution, local light intensity must be compensated. Unfortunately, this requires a precise map of tissue optical properties, which cannot be measured or calculated during imaging. Several methods have attempted to estimate and compensate fluence variations [34][35][36][37][38][39][40][41][42][43], but none work well clinically where real-time or near real-time corrections are needed. In most cases, fluence is compensated using an approximate exponential function equalizing intensities. This can help, but measured absorption spectra do not necessarily represent true molecular profiles. This is especially true for spectroscopic measurements at depth determining blood oxygenation or targeted nanoparticle concentration.
Second, tissue motion affects clinical spectroscopic PA imaging. For spectroscopy, the same object must be probed at several wavelengths. Measurements at each wavelength require a unique laser pulse, with multiple pulses producing a spectrum. The repetition rate is determined primarily by the maximum permissible exposure (MPE) into the body. The MPE is 20 mJ/cm 2 fluence and 1 W/cm 2 irradiance [44]. Thus, optimal signal to noise ratio (SNR) occurs at 50 Hz.
At least 5 wavelengths are needed for stable spectral decomposition [45]. Switching laser wavelength also takes significant time. As summarized in Fig.1d, no commercial systems run faster that 12 Hz for spectroscopy; using 5 wavelengths yields a 2 Hz effective frame rate.
Given typical physiologic motion, 2 Hz spectroscopic imaging has large artifacts. As illustrated in Fig.1b, motion changes the local concentration of absorbers over the spectroscopic sequence, resulting in inaccuracies at best and total destruction of the spectrum at tissue interfaces (see Results). To avoid blurred images and inaccurate spectroscopic data, scan rates should be increased and/or frames aligned with motion correction. Previous efforts include respiratory or data-driven gating [46,47], model-based estimation [48], and tissue boundary tracking (e.g., skin surface) [49]. Gating-based methods typically reject images during large motion, slowing the effective frame rate and limiting accuracy for fast processes. Correcting motion, rather than rejecting it, to preserve spectroscopic frame rates has not been demonstrated.
Here we introduce a different approach (see Fig.2) leveraging a unique diode-pumped wavelength tunable (700 nm -900 nm) laser emitting about 1 mJ pulses at 1000 Hz, with wavelength switching in less than 1 ms for any arbitrary wavelength order (i.e., wavelength need not be sequentially stepped between bounds) (Supplementary Note 1). Thus, every pulse in a sequence can be at a different wavelength without sacrificing repetition rate. To maximize exposure, we illuminate with a narrow (~ 1 mm in diameter) laser beam and switch it from fiberto-fiber at 1000 Hz around the US probe (Fig.2), resulting in one loop forming a singlewavelength frame in only 20 ms. The next loop uses another wavelength without delay; the procedure repeats over all wavelengths. In other words, instead of illuminating with a broad beam, we use fast-scanning (or fast-sweep) over the same illumination area.
The high-speed laser is extremely stable and externally triggerable at a variable rate. Thus, it can be integrated with conventional US sequences, enabling PA measurements interleaved with all US modes (e.g., harmonic and color flow imaging and elastography) at 50 Hz frame rates for both modalities (optical delivery details are in Supplementary Note 2; the specific scan protocol is in Results and Supplementary Note 3). In addition, we dramatically reduced the laser footprint and its cost. Most importantly, we will show that the fast-sweep concept has significant advantages over conventional broad-beam illumination because it enables simple methods for laser fluence correction and motion compensation (Results).

Results
Fast-sweep spectroscopic PAUS for laser fluence and motion correction. This method leverages recent developments in the laser industry. First, a diode-pumped wavelength tunable (700 nm -900 nm) laser was customized for fast-sweep imaging, in contrast to customizing the imaging system to the laser. It is compact (Supplementary Note 1), with a footprint potentially fitting within a cart-based ultrasound system. It emits pulses of about 1 mJ energy at a 1000 Hz repetition rate, with wavelength switching in less than 1 ms for arbitrary sequences. Thus, 1 kHz operation does not change between single wavelength and spectroscopic approaches. wavelength-tunable (700 -900 nm) diode-pumped laser (TiSon GSA, Laser-export, Russia), an integrated fiber delivery system (TEM-Messtechnik, Germany), and an US scanner (Vantage, Verasonics, WA, USA). The laser, externally triggered by the US scanner, emits pulses of about 1 mJ energy at a variable (up to 1 kHz) repetition rate, with wavelength switching times less than 1 ms for any arbitrary wavelength order. Thus, every laser pulse in a sequence can be at a different wavelength without sacrificing the kHz repetition rate. The integrated fiber delivery system includes 20 fibers arranged on the two sides of a transducer array. The spinning motor rotates the laser beam over the ring, thus sequentially coupling laser pulses to different fibers while sending a trigger signal to the US scanner. Upon receiving the trigger, the US system initiates the interleaved B and PA imaging sequence by sending a trigger to the laser. b, Timing diagram and pulse sequence for duplex interleaved, multispectral PAUS. Ten (10) wavelengths (i.e., 700, 715-875 nm every 20 nm) are used. For each wavelength, the laser beam irradiates tissue sequentially from 20 fibers, with several scanned, focused US beams interleaved for each laser firing. A frame composed of 1 B-mode and 20 PA sub-images is produced within 20 ms, i.e. the effective imaging frame rate is 50 Hz. c, Illustration on how high frame-rate B-mode US co-registered with PA images helps correct motion artifacts in PA images recorded at different wavelengths. The inter-frame tissue motion map is obtained using US speckle-tracking [50], then applied to all pixels of the co-registered PA images. Although US image coherence is quickly lost over a complete set of B-mode images in a sequence, tissue motion between adjacent B-mode frames can nearly always be tracked, accumulated over a sequence interval, and then applied to all PA images recorded at different wavelengths. Blue and yellow circles show local motion between adjacent frames, whereas a green circle fixed in location clearly shows the efficacy of motion correction. d, Light emerging from different fibers propagates different distances to a target located within tissue. The amplitude of a partial PA image obtained for single fiber irradiation follows the dependence shown in upper left plot. Considering the distance from each fiber to a typical absorber in the imaging field, the PA amplitude follows the form shown in upper right plot) due to light absorption and scattering in tissue. These measurements are used for robust estimation of the laser fluence in the medium independent of the details of the wavelength-dependent absorption curve of the specific absorber used for estimation. The procedure is repeated for all wavelengths used. When laser fluence is evaluated, it can be decoupled from the PA image to obtain the true light absorption spectrum for molecular absorbers.
Unlike previous delivery systems coupling laser pulses into all fibers in a bundle simultaneously [9], we couple light into individual fibers sequentially (see Figs.2a,b,d). Using a rotating wedge, the laser beam is projected onto a circle at the focus of a collimating lens. The wedge motor's absolute position controller synchronizes emission (i.e. coordinate of the laser spot on the circle) with the centers of 20 fibers in the bundle. With absolute position control, a precise rate is not needed for external laser triggering, ensuring maximal light delivery to each fiber. Motor speed variations only slightly alter the overall frame rate of 50 Hz.
Ten fibers are uniformly spaced along each elevational edge of the US imaging array (Fig. 2a). We integrated all controls, including laser pulse activation/sequencing, motor scanning, and PAUS image acquisition, with a commercial scanner (Vantage, Verasonics, WA, USA). The motor encoder triggers the US system to launch interleaved US and PA pulse sequences and the US system externally triggers the laser, synchronizing all sub-systems (Fig. 2a). Unlike triggering a scanner with a fixed rep-rate laser, externally triggering the laser with the scanner guarantees jitter-free synchronization by referencing both the imaging sequence and acquisition to the same clock. The scan protocol forming simultaneous PA and US images at a fixed wavelength is described in Methods.
To enable stable spectral decomposition, 10 laser wavelengths (i.e., 700, 715-875 nm every 20 nm) comprised the spectroscopic sequence. It can be customized in number of wavelengths, number of pulses per wavelength, wavelength sequence, and spectral resolution. Wavelength spacing is arbitrary, including a variable pitch, with 2 nm resolution defined by the spectral line width. For noise minimized spectral estimates, we turned off (0% energy) the laser at 700 nm to estimate noise levels (Methods).
Interleaved data acquisition provides simultaneous anatomic (US) and molecular (PA) images at a 50 Hz frame rate. This rate is sufficient for US speckle tracking [50] of individual pixels to map tissue motion between sequential images at different wavelengths (Fig. 2c). This motion can be compensated (Methods and Supplementary Note 9), as shown in the 3 rd column of Fig.2c, for all 10 wavelengths. After compensation, every pixel carries information from all wavelengths without motion artifacts and, therefore, enables spectral decomposition of molecular constituents.
The images in Fig. 2c are in vivo data from a small animal. Even at a 50 Hz spectroscopic frame rate, pixel displacements can be about a millimeter whereas the pixel size is less than 100 micrometers.
Even with no motion artifacts, the PA image amplitude is still proportional to the product of light absorption and laser fluence, where fluence is a function of depth and optical wavelength in biological tissue. Here, we use partial PA images from every fiber to estimate laser fluence.
Indeed, when light emerges from different fibers, it propagates different distances to a target. Fig.   2d (upper left plot) shows how PA signal amplitude changes with fiber index. Converting fiber index to fiber-absorber distance, PA signal loss with distance due to light attenuation is shown in Fig. 2d (upper right plot). Note that fluence losses with depth will differ for different wavelengths. As shown in Methods, such measurements can drive accurate and robust mapping of laser fluence independent of the wavelength-dependent absorption curve for a specific absorber. After fluence evaluation, it can be decoupled from the PA image to obtain the true light absorption spectrum of molecular absorbers. If fluence is ignored, accurate molecular imaging based on spectral decomposition is nearly impossible. We believe that this is especially true for imaging in humans, although we note that for mouse imaging fluence correction can sometimes be ignored.    Fig. 3d, Supplementary Figs.5a and b). The GNR spectrum is significantly red shifted. Moreover, the Higgins black ink spectrum is inverted from the ground truth; that is, its slope with respect to wavelength is the negative of the true slope. Note that these dramatic changes are at less than 1 cm depths, for effective light attenuation in the medium less than 3 cm -1 ; that is, under optical conditions typical in humans [33]. Unfortunately, this serious problem is usually omitted in the literature or not discussed in detail.
Leveraging the fast-sweep approach, we adopted a light diffusion model (Methods and

Ex vivo spectroscopic PAUS to guide interventional procedures
Real-time US is commonly used for interventional procedures [59][60][61], often guiding drug injections to help visualize the needle relative to anatomy and deliver the drug to the desired target. The drug itself cannot be visualized unless the injection creates bubbles. Such bubbles typically disappear quickly, and distribution of the drug is not always clear. Additionally, it takes great skill to orient the US imaging plane relative to the needle since a specular reflection is used for visualization. Nevertheless, real-time US guidance of many interventional procedures is a rapidly growing field that could expand greatly by overcoming these limitations.
PA guidance of needle injections has also been demonstrated [30,62,63]. Because the PA signal is quite independent of needle position relative to the transducer, precise orientation of the image plane is not needed, potentially making the technique more widely accessible. PA spectroscopic imaging can also add a molecular dimension because drugs can be molecularly labelled. Many small animal studies have shown the potential of spectroscopic PA molecular imaging [20][21][22]51]. Nevertheless, these methods have not translated well into clinical tools. Here, we demonstrate how fast-sweep PAUS provides robust molecular imaging for interventional procedures using a simple example of GNR injection ex vivo (chicken breast - Fig.4). This image-guided procedure has three sequential steps: (i) -needle insertion into tissue, (ii) -injection of a GNR solution, and (iii) needle pullout. A custom pulse sequence was developed (Fig.4 -left) ]. An alternate approach is to project the light absorption spectrum at every pixel onto the spectra of molecular constituents in tissue. This correlation-based method does not require numerical minimization (i.e. inversion), which is very sensitive to background absorption and noise [51,64]. It solves the forward problem, which, by definition, is more stable.
The upper row of Fig.4 shows PA images (after motion correction and fluence compensation) after full needle insertion but before injection. The Σ λ -PA image clearly shows the needle, but some additional bright spots are also present. It has high SNR because it coherently combines all 10 wavelengths over the spectral range; however, it is not specific to molecular constituents and contains artifacts. When spectrally projected to the GNR spectrum, the PA image shows nearly nothing over a 40 dB dynamic range. Indeed, nanoparticles had not been injected yet. The needle spectrum projection (Fig. 4 -upper right) clearly presents the needle with few artifacts. 14 The middle row of Fig.4b shows PA images after injection. Additional signals are evident in the Σ λ -PA image. Although the PA image differs greatly from that before injection, the B-mode image is nearly identical, demonstrating how poorly US monitors injections. Componentweighted PA images clearly differentiate the needle from GNR.
Finally, when the needle is removed (bottom row in Fig.4), the Σ λ -PA image is almost identical to that of the GNR-projected one (second and third columns, respectively), and no needle is observed. It is interesting that needle pullout leaves a trace of GNR in the needle channel.

In vivo spectroscopic PAUS in a small animal model
PA spectroscopic imaging has been extensively studied in small animal models [10,12,13,16,[20][21][22]51]. However, small animals greatly simplify imaging conditions. Light scattering is much lower than in humans and light easily penetrates the whole animal, especially when illuminated from all directions. Additionally, the transducer array can surround the animal, recording the PA signal with large spatial and temporal bandwidth for accurate PA reconstruction. Such conditions are very difficult to duplicate for most human applications, with breast as the notable exception.
In the last two sections, we addressed wavelength-dependent fluence variations. Here we use a small animal model to address tissue motion, the second major limitation on clinical translation. The specific mouse model is described in Methods. The GNR solution was injected into the mouse's right leg muscle using the same protocol described for ex vivo studies above. In particular, the laser pulse sequence was scanned at 775 nm during needle insertion, followed by incrementally sweeping 10 wavelengths over 10 cycles during GNR injection (Supplementary Movie 2).

Pixel-wise estimates of motion vectors from real-time US images (Methods and Supplementary
Note 9) show that motion differs from one pixel to another (Fig. 5b) and changes during the imaging sequence. Motion artifacts blur the Σ λ -PA image (top left panel in Fig. 5c). Furthermore, the needle is not removed from the GNR-weighted PA image (Fig. 5c - Fig. 5c) and the sensitivity of the GNR-weighted PA image is greatly improved, with more GNR particles clearly detected.
In another example, more GNR particles were injected for easy visualization. PA images at individual wavelengths are presented in Fig.6a, as well as motion-compensated Σ λ -PA (Fig.6b, left panel) and motion-compensated GNR-weighted (Fig.6b, right panel) images. Interestingly, not all bright points in the Σ λ -PA image appear in the GNR-weighted image.

Fig. 5: Case #1 of in vivo spectroscopic imaging for GNR injection into a mouse right leg muscle. Effects of motion artifacts on quantitative PA measurements are illustrated. a,
Experimental setup. b, Inter-wavelength motion artifacts during nanoparticle injection (See movements in B-mode images with wavelength). Inter-wavelength motion vector is estimated from two successive Bmode images using PatchMatch-based speckle tracking, and then accumulated over each wavelength sequence to produce the final motion-corrected PA images. For easy visualization, vectors are spatially decimated. c, The original uncompensated wavelength-compounded Σ λ -PAUS (left) and the corresponding GNR-weighted PA (right) images. Motion artifacts corrupt GNR-weighted PA images as the needle is not rejected. d, The needle is completely rejected in GNR-weighted PA images and GNR detectability is also improved in motion-corrected images.
After motion compensation, the measured GNR spectrum closely matches ground truth and, therefore, no fluence compensation is required. However, the correct spectrum cannot be

Discussion
Spectroscopic PA imaging systems using bulky solid-state lasers have not translated well clinically for many reasons, the most fundamental being they do not robustly correct fluence or compensate motion. Practically, their size and cost also limit easy integration with clinical US. Laser fluence compensation is key to fast-sweep PAUS. The PA signal is proportional to both the local light absorption coefficient and laser fluence. Although unnecessary for some animal models, in vivo human measurements require fluence compensation for PA spectroscopy.
Without it, PA image spectra can be markedly different from true spectra (Fig. 3). Indeed, for the model system presented here, the GNR spectrum was significantly red shifted and the black ink spectrum slope even changed sign. This can lead to erroneous conclusions about molecular contributors to the PA signal. With it, spectroscopic PA images can be decoupled from wavelength-dependent fluence variations, helping identify molecular constituents based on known optical absorption spectra (Figs. 3 and 4). Because all pixels with amplitudes exceeding the noise floor can contribute to fluence estimates, this procedure is almost guaranteed to be stable for optically quasi-homogeneous media (Supplementary Notes 6-8 for details).
For over twenty-five years, the nearly unique properties of US speckle have been exploited for dense estimates (i.e., full displacement vector at every pixel) of tissue motion. Using a speckletracking algorithm appropriate for real-time use, dense displacement fields were estimated from interleaved US images at the 50 Hz frame rate. Since both modalities use the same array, US and PA pixels are co-registered. Thus, US-derived displacements can correct PA images for interframe motion (see Supplementary Note 9), aligning PA images from all wavelengths in a sequence.
For our 10-wavelength sequence, motion artifacts are clearly very serious in vivo (Fig. 5). For high spatial resolution (i.e., pixel-wise) spectroscopic imaging, motion corrupts spectral measurements, which cannot always be recovered using spatial averaging. For instance, detecting multiple molecular constituents or separating exogenous agents (like moleculartargeted nanoparticles) from endogenous absorbers (like blood) is challenging if motion is not properly corrected.
Motion compensation may also help tackle limited PA penetration (and, therefore, typical low PA SNR). Without considering motion, signal averaging will not significantly increase SNR and will spatially blur spectroscopic information. With it, however, multiple frames can be averaged to greatly enhance SNR and increase image depth. In addition, motion compensation may be very important for fast processes, as often encountered in interventional procedures. Although motion artifacts can sometimes be rejected in small animal models [46], they must be considered in clinical imaging.
The current system has been reprogrammed to produce US images at a 5 kHz frame rate (plane wave imaging [76]) but with image quality markedly reduced from the current approach. Hybrid sequences can be developed to trade off image quality with frame rate, providing robust tracking for any significant physiologic motion. If faster rates are needed, then dense motion fields can be interpolated to any time and space point to compensate motion, even PA sub-images acquired at the same wavelength but with different fibers.
Building on motion correction and fluence compensation results, we proposed two PA modalities: spectral inversion of all known molecular absorbers in the medium. It may be unstable even for fluence compensated PA images due to typically low SNR and image artifacts. In contrast, we use the projection of the measured absorption spectrum to that of a known component. We used component-weighted imaging to identify GNR and a needle (see Fig.4). For multiple molecular constituents, it can be performed for every constituent. Because spectral projection is correlationbased and does not use inversion, we believe that it can be more stable. We note, however, that projections may not yield absolute constituent concentrations. Possibly both methods can be combined whenever absolute concentrations are needed.
Although fast-sweep spectroscopic PA imaging has significant advantages, it also has limitations.
As noted, the probe's limited view and finite bandwidth produce image artifacts, especially for large objects with uniform absorption [56,66]. In addition, the small footprint of individual laser firings reduces SNR compared to broad illumination. For the sequence used here, the SNR is reduced approximately by the square root of the number of fibers. That is, the current system has approximately 13 dB lower SNR compared to broad illumination with a 50 Hz high-power laser delivering the same surface fluence. Because of our high frame rates and laser stability, however, SNR can be recovered with averaging. For example, a Σ λ -PA image can recover nearly 9 dB.
With motion compensation, longer averaging periods can also improve SNR.
Given its advantages and limitations, real-time spectroscopic PAUS imaging is appropriate for many clinical applications but is not appropriate for some, such as deep imaging within relatively high scattering tissue where SNR is a significant concern. There are many potential clinical applications, but two obvious short-term targets are guiding interventional procedures, such as the needle injections presented here, and monitoring the patency of full thickness skin grafts.
US-guided needle-based procedures are challenging because alignment is difficult, agents delivered through the needle cannot be visualized, and US contrast agents delivered simultaneously are short-lived and can only help confirm the delivery site. On the other hand, both drugs and cells can be labelled with FDA-approved PA contrast agents such as Indocyanine green (ICG) and methylene blue (and many others) for procedure guidance. These agents are molecular, so they persist for long periods (hours to days) to help monitor drug/cell migration.
Thus, real-time spectroscopic PAUS imaging can not only guide drug/cell delivery, but also monitor diffusion and migration over long periods and correlate movement with outcomes.
Real-time spectroscopic PAUS imaging can also potentially assess longitudinally the patency of Both B-mode and PA images were formed using coherent delay-and-sum beamforming, followed by envelope detection. The flexibility of our pulse sequence enables multi-beam acquisition (dual receive beams in this study) to maintain high US frame rates. By interleaving laser firings with US pulse sequences, PA imaging can also be combined with other US modalities, such as color flow and harmonic imaging and real-time elastography.
In addition to US B-mode images, we produced wavelength-compounded (Σλ-PA) and Estimated displacements are applied to each wavelength PA image before fluence compensation.
The specific motion compensation scheme here used US images acquired at a 50 Hz frame rate.
This was sufficient to track physiologic motion in the present study. It may be insufficient, however, for other applications including faster motion, especially near large pulsatile vessels.
Here, laser fluence is evaluated automatically during PAUS imaging without additional equipment and delays. Because tissue illumination (see Fig.2d) is performed sequentially with 20 individual fibers to form 20 partial PA images, the local amplitude of partial images is a function of fiber index, i.e. the distance between the fiber source and target (see Fig.2d). The PA image contains multiple individual pixels and, therefore, the amplitude dependence on distance between any pixel and the source can be obtained for many points with partial PA image amplitudes over the noise floor. These measurements provide inputs to fluence reconstruction.    Fig. 5d). The solution used in Fig. 3 contained 0.47 ml Prussian blue nanoparticles mixed with 380 ml de-ionized water and 20 ml 20% Intralipid.