The brain is the information-processing unit of the human body and the most important part responsible for human cognition including learning, memory, and thinking. Meanwhile, it is the most complex thing not yet fully discovered, regarded as the last and grandest biological frontier [1–4]. With the development of advanced brain imaging tools such as functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and magnetoencephalography (MEG), it is possible to noninvasively visualize and quantify the brain structure, function, and metabolism and thus better understand and diagnose brain tumors or neurological diseases such as Alzheimer’s disease [5–7]. However, these techniques either suffer from radioactive hazards or a low spatial resolution on the millimeter scale, hindering the capture of subtle brain anatomical or functional abnormalities for early disease diagnosis and regular examination for effective treatments without the risk of overdose radiation [8,9]. Photoacoustic tomography (PAT) is a noninvasive and hybrid biological imaging technique that integrates the advantages of pure optical imaging and ultrasound imaging. Relying on the detection of optically excited acoustic waves or photoacoustic (PA) waves, PAT with high optical contrast and large penetration depth can provide rich anatomical and functional information of biological tissues and serve as a useful tool for vascular angiogenesis and biomolecule-tagged tumor imaging in biomedical research or clinical diagnosis [10–13].

In general, PAT can be divided into two types: photoacoustic microscopy (PAM) and photoacoustic computed tomography (PACT). Compared to the PAM, which uses focused light or ultrasound to acquire images through point-by-point raster scanning, PACT employs diffusive light to illuminate the whole imaging region. The thermally excited ultrasound from the tissue after absorbing the short laser pulses then propagates through the tissue and reaches the ultrasound transducer. Two-dimensional (2-D) or three-dimensional (3-D) images of the tissues can be reconstructed by using the received ultrasound signals at different locations [14–16]. During the imaging process, the photoacoustic signals carrying the time-of-flight information of the biological tissues are routinely detected by the piezoelectric (PZT) ultrasound transducers [17–19]. However, due to the size-dependent sensitivity of PZT transducers, the majority of the PZT transducers in the PAT systems have aperture sizes on millimeter to centimeter scale to obtain sufficient sensitivity. One problem existing for these large-sized PZT transducers is the limited acceptance angle, which inevitably causes artifacts, information loss, and resolution degradation to the images. This limited-view issue induces degradation in the image quality and is even more severe for high-frequency imaging systems involving a shorter ultrasound wavelength [20]. Ideally, point ultrasound detectors with small active areas have large angles of acceptance, but their sensitivity is severely degraded by high thermal noise [21]. Therefore, without greatly compromising the sensitivity, the focus of the transducer has been utilized as a virtual point detector to increase the acceptance angle and thus the spatial resolution [22]. This concept is further implemented on a concave PZT transducer with dual foci and a flat transducer attached to the convex acoustic lens to form focusing [23]. By using a customized convex transducer to form negative focus, Nie et al. demonstrated high-quality imaging of the cerebral cortex of a monkey brain with improved spatial resolution and field of view (FOV) [24]. However, attachment of an acoustic lens to a flat transducer suffers from the acoustic impedance mismatch between the lens and transducer material, which induces ultrasound loss and reverberation. The concave or convex transducers increase the fabrication complexity and thus require customization [25].

Optical ultrasound transducers such as a micro-ring resonator, silicon waveguide, and polymer Fabry–Perot (F-P) cavity exhibit high sensitivity and small form factors, providing promising solutions to sensitive ultrasound detection with wide acceptance angles [26–29]. The micro-ring resonator and silicon waveguide transducers achieve a broad frequency bandwidth of up to 100 MHz, which can greatly increase the spatial resolution. The on-chip structure also allows easy design and manufacture of large-scale transducer arrays, which is attractive to increase the imaging speed for real-time imaging of fast-moving objects such as live zebrafish, whereas the limited acceptance angle of ∼60deg degrades the lateral spatial resolution in linear scanning PACT imaging applications. Notably, a fiber-tip ultrasound transducer with the convex polymer F-P cavity coated with high-reflectivity dielectric film demonstrates a low noise equivalent pressure (NEP) down to 2.6 Pa and a nearly omnidirectional ultrasound response. The wide acceptance angle of ∼180deg enables large FOV for raster-scanning optical-resolution PAM. However, the F-P cavity with a high optical finesse features a sharp notch in its reflection spectrum, thus requiring sophisticated locking of the narrowband probe laser at the spectral slope for optimal signal output. In addition, the F-P cavity without focusing suffers a rapid amplitude decay for the diverging photoacoustic waves generated by the biological tissues [30].

Herein, we propose a linear-scanning PACT system based on a negatively focused fiber-laser ultrasound transducer, capable of noninvasive imaging of deep mouse brain with high spatial resolution. By exploiting the fiber flexibility, the fiber laser is mechanically bent into a convex shape, which forms a negative ultrasound focus with no need of complex processing or extra acoustic lens. A centimeter-long fiber laser with stable output is prepared by optimizing the inscription parameters, which not only increases the acceptance angle of the transducer to nearly 120 deg but also reduces the noise level and thus improves the signal-to-noise ratio (SNR) for detecting weak PA signals from deep tissues. A synthetic aperture focusing technique (SAFT) is applied for image reconstruction by treating this convex fiber-laser ultrasound transducer as the virtual point detector. The built linear-scanning PACT system demonstrates in vivo imaging of intact mouse brain with a penetration larger than 7 mm and a high spatial resolution of ∼130μm. The system is further applied for the study of a mouse model with intracerebral hemorrhage, which shows its great potential for biomedical research and clinical diagnosis.

In PAT, the photoacoustic excitation by a nanosecond (ns) pulsed laser satisfies the thermal confinement under the assumption that the acoustic medium is lossless, dispersion-free, and homogeneous [31]. The 3-D pressure field p(r,t) fulfills the wave equation ∂2p(r,t)∂t2−c2∇2p(r,t)=ΓHr(r)∂Ht(t)∂t,(1)where p(r,t) is the generated pressure, t is time, r=(x,y,z) is the position in 3-D space, c is the acoustic velocity in the medium, Γ is the Grüneisen parameter, and Hr(r)Ht(t) is the energy per unit volume and unit time. For a point source irradiated by a short laser pulse, there are approximately Hr(r)=δ(r−r′) and Ht(t)=δ(t). In frequency domain, there is Ht(t)=exp(−iωt) for a point source. The pressure field can then be expressed as pδ(r,k)=−iΓk4πexp(ik|r−r′|)|r−r′|,(2)where k is the acoustic wave number given as k=ω/c.

Here, the convex fiber-laser transducer is treated as a uniform 2-D arc detector to simply consider the effect of its geometry on the PA response as shown in Fig. 1(a). The arc detector with a curvature of radius of R locates in the plane at z=0 and has a distance of D to the point PA source. If k|L(θ1)−L(0)|, k|L(θ2)−L(0)|≫1 and θ1<0<θ2, the expression for the frequency-domain response of the arc detector can be simplified using the stationary-phase method (SPM) as [32] pD(0,k)≈DRD+R2πkpδ(D,k)exp(iπ4),(3)where pδ(r,k) is defined by Eq. (2), with r replaced by D here. Based on the hypothetic condition for deriving Eq. (3), in the high-frequency regime only the PA waves propagating along the direction of θ=0, or with the wavefront tangentially intersecting with the arc detector at a geometrical point within the arc region, can contribute to the signals detected by the arc detector [33]. Otherwise, the high-frequency spherical PA waves at different positions along the arc detector cancel each other due to the phase mismatch, resulting in severe signal loss. In addition, Eq. (3) suggests that there is a low-pass filtering effect associated with k−1/2 for the arc detector.

To further examine this effect, the responses of the arc detector and a point detector to the same point PA source are simulated using the k-wave MATLAB toolbox, as shown in Fig. 1(a). The point detector locates at the focus of the arc detector with a curvature radius of 15 mm and the point source with a radius of 25 μm locates at a distance of 30 mm from the point detector. The central frequency and bandwidth for the point detector and the convex fiber detector are set the same, with the values of 15 MHz and 100%, respectively, close to the actual values of the fiber laser as characterized in the experiment.

As shown in Fig. 1(b), the time-domain PA signals detected by the two detectors of the same limited bandwidth and the corresponding frequency spectrum after the Fourier transforms are shown in the inset of Fig. 1(b). Here, the positive peak amplitudes of the time-domain PA signals are normalized and the time delays of the two signals are shifted to reduce their time interval for clear comparison. There is slight attenuation in the high-frequency component of the PA signal for the arc detector, which seems to deviate from that predicted by Eq. (3). This is a result of the limited bandwidth of the PA wave emitted from the point source with a fixed diameter and the limited detector bandwidth. By dividing the response of the arc detector with that of the point detector, the effect of the above factors on the detector frequency response can be excluded. The ratios of the frequency responses of the two detectors for the point source with different radii are plotted in Fig. 1(c). By taking the logarithm, the line slope of −1/2 as shown in the inset of Fig. 1(c) denotes a low-pass filtering effect exactly the same with the relationship of k−1/2 as indicated by Eq. (3). Despite the low-pass filtering effect, the fiber-laser transducer used in the study has a limited bandwidth similar to that of most PZT transducers, making this effect negligible for the imaging results.

The photograph of the negatively focused fiber-laser ultrasound transducer is shown in the inset of Fig. 1(d). Two fiber Bragg gratings are inscribed in a piece of Er/Yb-doped fiber using 193-nm UV light and a 1063.86-nm phase mask to form an inline Fabry–Perot (F-P) cavity with the reflection spectrum centered at 1542.8 nm. By reducing the modulation depth of the refractive index in the Er/Yb-doped fiber and simultaneously increasing the grating period number, a fiber laser cavity with an effective cavity length of ∼30mm and stable laser output is obtained. Single-mode fibers (SMFs) are fusion spliced to each end of the doped fiber. The fiber laser is bent into an arc shape with a curvature radius of 15 mm and fixed by a lab-designed holder. As the 980-nm pump light enters the fiber, the two orthogonal polarization modes (fx and fy) near 1543 nm as a result of the intrinsic fiber birefringence [see Fig. 1(e)] generate a radio-frequency beat signal (fb=|fx−fy|). When ultrasound propagates to the fiber laser, the stress-induced birefringence leads to frequency shifts of the signals that can be read out by an I/Q phase demodulation unit. By increasing the effective length of the laser cavity to 30 mm, the acceptance angle can be calculated to be ∼116deg. In addition, the output signals of the 30-mm-long fiber laser exhibit a low noise level (3σ) down to 12.3 kHz, as a result of the reduced thermal noise [34]. As shown in Fig. 1(f), the noise level is over twofold lower than that of a 6-mm-long laser cavity. Based on the measured ultrasound sensitivity of 2.25 MHz/kPa of the fiber laser [35], the NEP is approximately 5.4 Pa within a detection bandwidth of 50 MHz. Figure 1(g) shows the PA signal detected by the fiber-laser ultrasound transducer. By performing the Fourier transform on the time-domain PA signal, the corresponding frequency response is shown in the inset of Fig. 1(g). The transducer bandwidth is ∼20MHz and the center frequency locates at ∼15MHz.

Figure 2(a) is the schematic of the PACT system using the negatively focused fiber-laser ultrasound transducer. The pulsed light is delivered and irradiated onto the imaging objects through a fiber bundle. Pump light with an optical power of ∼400mW from a 980-nm laser module (HYLM-980-CW, Tianjin UniStarCom Technology) is fed into the fiber laser through a wavelength division multiplexer (WDM). The synchronous clock signal for triggering the pulsed laser, translation stage, and data acquisition (DAQ) unit is provided by a digital I/O board (PXIe5641, ART Technology) controlled by a LabVIEW program. The fiber-laser ultrasound transducer is linearly scanned by the stage along the x-axis and the PA signals generated at each step are detected by the fiber-laser ultrasound transducer. The optical signals output from the fiber laser travel through the WDM and the polarizer and then reach the photodetector (PDR-10-S-FA-K, Realphoton). The output electrical signals from the photodetector are converted to time-domain PA signals by the I/Q demodulation unit and then acquired by the DAQ for subsequent image reconstruction.

As shown in Fig. 2(b), the fiber-laser ultrasound transducer in a convex shape forms a negative focus at the arc center, which is treated as a virtual detector here. This virtual detector concept, in combination with the SAFT, can effectively increase the view angle and thus improve imaging quality [36]. The virtual detector actually superimposes a synthetic aperture on the PA sources in the overlap region at all adjacent positions during the imaging process.

To study the acceptance angle of the fiber-laser transducer, its acoustic spatial response has been simulated using the k-wave MATLAB toolbox, with the result shown in Fig. 3(a). The measured spatial response agrees well with the simulated result as shown in Fig. 3(b) (see Appendix A.1). The green arc in Fig. 3(b) represents the location of the fiber-laser ultrasound transducer. At the vicinity of the fiber, the ultrasound response is missing to avoid the contact of the fiber-laser transducer with the hair during raster scanning. The estimated acceptance angle from Fig. 3(b) is ∼116deg. The acceptance angle can be further increased by reducing the bending curvature or increasing the cavity length of the fiber laser. Currently, the minimum curvature radius achieved is ∼15mm as limited by the mechanical strength of the fiber laser. To increase the cavity length, parameters of the grating inscription including the grating length and refractive index modulation depth are optimized to obtain a 30-mm laser cavity without mode hopping during the imaging process. Further increasing the cavity length will deteriorate the stability of the laser output due to the multiple-longitudinal-mode effect. A potential solution to obtain a longer laser cavity is to adopt an additional π-phase-shift FBG as the filter or construct a composite cavity [37,38], which is under investigation.

The acquired PA image of the hair loop is shown in Fig. 3(c), which agrees with the contour of the hair loop in the photograph as given in the inset of Fig. 3(c). Currently, the major part of the hair loop can be clearly observed but part of the information along the z-axis direction is lost since the fiber-laser transducer is still not omnidirectional. On the other hand, the lost information verifies the importance of increasing the acceptance angle to truly reflect the exact structure information of the target. The PA image of the leaf is shown in Fig. 3(e), similar to the structure observed from the photograph as shown in Fig. 3(d). The loss of the vertical leaf structure and the limited-view-related artifacts in the reconstructed image are also caused by the non-omnidirectional response of the fiber-laser transducer, similar to that for the hair loop.

The tangential and axial resolutions of the PACT system are estimated by imaging three 10-μm-diameter carbon fibers aligned along the y-axis [see Fig. 2(a) lower left and Appendix A.1]. The resolution is defined as the full width at half maximum (FWHM) of the carbon fiber profiles along the dashed lines as indicated in the reconstructed PA images as shown in Fig. 4(a). Unipolar images are obtained by applying the Hilbert transform during the image reconstruction. The original image, the image after Gaussian-window apodization, and the image after coherence-factor weighting are compared. It can be observed that the image using coherence-factor weighting yields minimal artifacts and the image using Gaussian window apodization can also efficiently reduce the artifacts. The estimated tangential resolution and the axial resolution are ∼122μm and ∼123μm, respectively, attributed to the broad bandwidth and large acceptance angle of the fiber-laser ultrasound transducer. Despite that coherence-factor weighting can improve the tangential resolution and reduce the reconstruction artifacts for imaging the carbon-fiber phantom, it needs to be mentioned that the effectiveness significantly reduces in complex biological tissues such as mouse brains. Therefore, for the deep-mouse-brain imaging, Gaussian-window apodization instead of the coherence-factor weighting is adopted.

The linear-scanning PACT system is used to image the brains of BALB/C mice (6 weeks old, Charles River) in vivo (see Appendix A.2). The acquired coronal sectional images of the mouse brain of two different locations are shown in Fig. 5. The scalp, the skull, and the whole brain including the cerebral cortex and thalamus (or the mesencephalon) can be clearly observed over a penetration depth of ∼7mm. In addition to the major brain structures, small vessels in the scalp and the cerebral cortex can also be resolved. The in vivo imaging of an intact mouse with large depth and fine resolution proves the potential of the linear-scanning PACT system for future small animal imaging and clinical diagnosis.

As illustrated in Fig. 6(a), by manually limiting the acceptance angle of the fiber-laser ultrasound transducer during the image reconstruction process, i.e., the PA signals acquired beyond the acceptance angle α are abandoned, the reconstructed images under different acceptance angles are obtained and compared in Fig. 6(b). For the small acceptance angle of 30 deg, only the structures horizontal or parallel to the scanning direction of the transducer can be captured. With the increasing of the acceptance angle, vertical structures such as the azygos pericallosal artery (azPA), rostral rhinal vein (RRV), and anterior choroidal artery (AchA) become clear, emphasizing the importance of the enlarged acceptance angle of the fiber-laser ultrasound transducer. As the mouse head in the current experiment is kept intact, the received PA signals and thus the image contrast can be further enhanced by removing the scalp and the skull in the future [39].

To demonstrate the potential of the system for biomedical applications, a mouse model of intracerebral hemorrhage is built by the blood infusion [40], with the process as sketched in Fig. 7(a) (see Appendix A.2). To make sure the hemorrhage is induced by the blood infusion at the designated location, a mouse with the brain inserted with a black-ink-coated optical fiber having a diameter of 125 μm is imaged first. The images before and after the fiber insertion are shown in Figs. 7(b) and 7(c). Thanks to the large acceptance angle, a clear outline of the inserted fiber can be observed, which shows the potential for needle guidance in biomedical research or clinical applications. Following the same insertion depth of ∼6mm and orientation, the intracerebral hemorrhage is induced into the mouse brain.

Figures 7(d) and 7(e) show the brain images of the same mouse with and without thrombus. For the mouse model with intracerebral hemorrhage, the PA signals boost up in the region of the rostral rhinal veins on the blood infusion side (the left side of the brain), with the amplitude two to four times greater than that without hemorrhage at the right side, verifying the high sensitivity of the system of detecting intracerebral hemorrhage. The large acceptance angle of the fiber-laser transducer also enables clear visualization of the thrombus region instead of simply the horizontal part similar to the case of α=30deg or 60 deg in Fig. 6(b). For the optical fiber insertion, no obvious brain bleeding is observed from the brain images [Figs. 7(b) and 7(c)]. To examine whether bleeding occurs for the syringe needle with a much larger diameter than fiber, the same procedure as that for constructing the mouse model of intracerebral hemorrhage is performed except that no blood is injected into the brain. The results are shown in Figs. 7(f) and 7(g), verifying the effectiveness of the model of the intracerebral hemorrhage. The image with clearly observed brain hemorrhage in Fig. 7(e) proves the ability of the PACT system for visualizing intracerebral thrombi and the potential for medical screening and diagnosis of stroke.

In this work, we have demonstrated a PACT system based on the negatively focused fiber-laser ultrasound transducer that enables deep noninvasive imaging of the mouse brains. The transducer in an arc shape forming a negative ultrasound focus exhibits an acceptance angle of nearly 120 deg, which enables a nearly isotropic spatial resolution of ∼130μm and noninvasive imaging of the whole mouse brain over a depth of 7 mm. By optimizing the inscription parameters, a centimeter-long fiber laser with stable output is prepared, which not only increases the acceptance angle but also reduces the noise level and thus achieves a low ultrasound detection limit of ∼5.4Pa. In vivo imaging of a mouse model with intracerebral hemorrhage is also showcased to demonstrate its potential for needle-guiding and stroke diagnosis. Currently, due to the lack of elevational focus and an acceptance angle of ∼60deg in the elevational direction [41], the out-of-plane signals would also be received and thus induce artifacts into the image for the cross-sectional brain imaging. Encapsulating the transducer between two plates coated with ultrasound-absorptive materials can potentially diminish the out-of-plane PA signals [42]. On the other hand, the ability to receive out-of-plane PA signals can be utilized to achieve volumetric imaging of the whole mouse brain by raster-scanning the transducer along both the lateral and elevational directions, despite that the 2-D scanning process and 3-D imaging reconstruction may be time-consuming. One potential direction is to increase the number of the fiber-laser ultrasound transducers and parallel demodulation of the output radiofrequency signals from the transducer array [27,43]. The transducer array can greatly reduce the imaging time but inevitably increase the cost. Overall, the system with high spatial resolution and large tissue penetration is promising for biomedical study and medical diagnosis such as early detection of stroke or tumors.

Acknowledgment. We thank A. P. Cheng Ma and Dr. Handi Deng in Tsinghua University for helpful discussions.