Photoacoustic imaging of visually evoked cortical and subcortical hemodynamic activity in mouse brain: feasibility study with piezoelectric and capacitive micromachined ultrasonic transducer (CMUT) arrays

This study investigates the feasibility of capturing visually evoked hemodynamic responses in the mouse brain using photoacoustic tomography (PAT) and ultrasound (US) dual-modality imaging. A commercial piezoelectric transducer array and a capacitive micromachined ultrasonic transducer (CMUT) array were compared using a programmable PAT-US imaging system. The system resolution was measured by imaging phantoms. We also tested the ability of the system to capture visually evoked hemodynamic responses in the superior colliculus as well as the primary visual cortex in wild-type mice. Results show that the piezoelectric transducer array and the CMUT array exhibit comparable imaging performance, and both arrays can capture visually evoked hemodynamic responses in subcortical as well as cortical regions of the mouse brain.


Introduction
Optical modalities have a wide range of applications in brain imaging and neuroscience research due to their excellent sensitivity to hemodynamics and spatial resolution in superficial tissues [1].However, the penetration and resolution of brain imaging are limited by optical scattering through the scalp, skull, and brain tissue.Photoacoustic tomography (PAT) is a modality that can capture optical contrast distribution in the brain with acoustic resolution and penetration [2,3].Relying on ultrasound (US) imaging systems for signal acquisition, PAT naturally possesses US imaging capability.Combining the anatomies in US imaging and local hemodynamics in PAT imaging, PAT-US parallel imaging shows great promise in quantifying brain activity in small animals.While PAT-US brain imaging studies have focused on the cerebral cortex [4][5][6][7], considering the 100-µm resolution at a penetration depth of a few centimeters [2], PAT-US imaging is theoretically capable of resolving the subcortical structures in small animals, where all brain nuclei are <1 cm from the skull surface.This ability to study the brain comprehensively and rapidly would expedite the characterization of mice and other species with comparable brain sizes.
Another advantage of PAT-US parallel imaging is the potential for miniaturization.Given the off-target effects of anesthesia and restraints [8,9], imaging free-moving animals with a lightweight, flexible, and wearable transducer array would allow functional assessment under normal physiologic conditions as well as facilitate longitudinal studies.Conventional US imaging utilizes bulky, rigid, and heavy piezoelectric transducer arrays.Thanks to the thin substrates used in fabrication, the nonnecessity of thick backing layers, and the ability to closely integrate transducer arrays with frontend electronics, capacitive micromachined ultrasonic transducer (CMUT) arrays present a superior way to miniaturize probes for wearables.Further, the high sensitivity due to the close integration of frontend electronics and the wide bandwidth of CMUTs yield high-quality images with a high signal-to-noise ratio (SNR) and fine resolution.We previously developed lightweight 1-dimensional (1D) [10] and 2D [11] wearable CMUT arrays for mice on flexible circuit boards with wireless communication or flexible cable connections to backend electronics.CMUT arrays have successful applications in Doppler velocity measurement [12,13], intravascular US [14], PAT [15,16] and US therapy [17][18][19].
We recently achieved PAT imaging of the visually evoked hemodynamic responses in the mouse visual cortex using a ring-shaped piezoelectric transducer array [3].In the present study, we evaluate the feasibility of using CMUT and piezoelectric arrays in PAT-US parallel imaging of subcortical as well as cortical hemodynamic activity in mice.The performance of the imaging system was first assessed in phantom studies.We subsequently examined the system's ability to capture visually evoked hemodynamic responses in the primary visual cortex (V1) and the subcortical superior colliculus (SC).

Imaging phantom
200-µm-diameter fishing lines were used to test the spatial resolution of the transducer arrays at various depths.The lines were submerged in water at an angle that crossed the imaging plane.The distance between the fishing line and the array surface was adjusted using a mechanical translation stage.The imaging resolution was determined by deconvolving the profiles of the fishing lines in the PAT and US images with the actual cross-section of the fishing lines, i.e., a circle with a diameter of 200 µm.

Animal procedures
All procedures were approved by the Institutional Animal Care and Use Committee at the University of Michigan.4-to 7-month-old male C57BL/6 wild-type mice (JAX000664, Jackson Laboratory) were used.Each array was examined using 3 mice.The mice were housed in a 12-hour light/12-hour dark cycle, and PAT imaging was conducted during the light phase.Prior to imaging, each mouse was dark-adapted overnight.Under dim red light and isoflurane (1-2%) anesthesia, the animal's scalp was removed, and the skull was slightly thinned to reduce optical and acoustic attenuation.The mouse was then secured onto a 37°C warming restraint attached to a multi-axis translation stage, and the cortex surface was aligned with the imaging plane.Following intraperitoneal injection of the sedative acepromazine (5 mg/kg), the mice were anesthetized by 1% isoflurane at an airflow rate of 1.5 L/min throughout the imaging procedure [20].
The visual stimulus was broadband white light generated by a fiber-optic halogen illuminator (HL150-B, AmScope) flickering at 1 Hz and positioned about 5 cm away from each eye, producing an irradiance of approximately 16 mW/cm 2 at the cornea.Before retinal photostimulation, the mouse was kept in darkness for 10 min for a baseline measurement.Then, both eyes were simultaneously exposed to the flickering stimulus for 10 sec, followed by another 10 min in darkness, during which PAT imaging continued to monitor hemodynamic recovery from the visual stimulation.Shortly afterwards, the animals were euthanized by carbon dioxide and cervical dislocation.

Transducer arrays and US platform
This study compared a piezoelectric transducer array with a CMUT array.Both were 1D arrays with 256 elements.The piezoelectric transducer was a commercial array (L8-18i, GE Healthcare) with a central frequency of 13 MHz and a -3 dB bandwidth from 8 to 18 MHz.The footprint of the array was 11.1 mm × 34.8 mm.The CMUT array was custom-made with a central frequency of 14 MHz and a -3 dB bandwidth extending from 3 to 25 MHz.The footprint of the array was 3.0 mm × 24.9 mm, as shown in Fig. 1(a) [21][22][23].The lateral and elevational dimensions of the elements are 99 µm and 2988 µm, respectively.The element pitch is 97 µm.Figure 1(b) shows the 3D printed holder integrating the transducer arrays and the illumination optical fibers for PAT. Figure 1(c) shows a schematic of the PAT-US imaging system.In the PAT mode, an Nd:YAG laser pumped optical parametric oscillator (OPO) (Phocus Mobile, Opotek Inc., Carlsbad, CA) was used as the excitation source.The output of the OPO had a pulse duration of 4-6 ns, a pulse repetition rate of 10 Hz, and a pulse energy variation of less than 5%.To ensure optimal illumination in the phantom study, we illuminated the fishing line with the original 5-mm-diameter fiber bundle of the OPO laser.The emission surface was positioned 30 mm from the fishing line.The fishing line was positioned 10, 15 or 20 mm away from the probe surface, with a US image and a PAT image taken at each position.The signals were not averaged.
For the animal study, before data acquisition, the transducer array was oriented in the coronal plane of the mouse brain and scanned across the entire brain.A plane covering both V1 and the SC was identified by comparing the anatomies in US images with the Allen brain atlas (Allen Institute for Brain Sciences, Seattle, WA).The PAT and US images were continuously recorded during the animal procedure described above.
For the PAT mode, the laser energy was delivered through a customized multimode fiber bundle with fiber tips attached on both sides of the linear arrays.The fiber bundle consisted of 32 fibers distributed in the 25-mm range on each side of the arrays.The projections of the fibers on both sides of the arrays overlapped at the mouse skull surface, which was 15 mm from the array surface, as illustrated in Fig. 1.All experiments used imaging illumination at 797 nm because mouse retinal photoreceptors are insensitive to near-infrared wavelengths, so the imaging illumination would not interfere with the visual stimulation.Moreover, at this wavelength, the oxygenated and deoxygenated hemoglobin had the same absorption coefficients [24].The maximum optical fluence at the skull surface was ∼20 mJ/cm 2 , below the American National Standards Institute safety limit.The photoacoustic (PA) signals were digitalized and sampled at 40 MHz using a Vantage 256 ultrasound system (Verasonics, Redmond, WA) with identical configurations for both transducer arrays.As determined by the laser repetition rate, the frame rate of the system was 10 Hz.
After each PA data acquisition, a US frame was acquired.Plane wave transmission was used with both arrays.Both US and PAT frames were acquired for 20 sec before the retinal photostimulation, 10 sec during the photostimulation, and 30 sec after the photostimulation, producing a total of 600 frames.Each mouse received just one photostimulation to ensure that all mice were tested under fully dark-adapted conditions.

Image reconstruction and signal processing
Both the US and PAT frames were reconstructed using delay-and-sum beamforming algorithms.
The process of extracting the responses in the visual regions is detailed below: (a) In the PAT image, a temporal trace was extracted at each pixel in the consecutive frames.
(b) Each temporal trace was first detrended by subtracting the linear fitting line of the pre-stimulation temporal trace to remove the systematic shift from the detected signal.
(c) The detrended temporal trace was normalized by the root-mean-square of the signal strength of the original pre-stimulation temporal trace before subtraction, resulting in the baseline-subtracted and normalized PA signals (∆PA/PA).
(d) A spatial moving average of 3 × 3 pixels and a temporal forward-moving average of 25 frames (2.5 sec) were also applied to filter out the noise stemming from random fluctuations.The spatial and temporal averaging size was selected following a pioneering study in optoacoustic imaging of the mouse brain [25].
(e) V1 and SC regions in the PAT images were identified by comparing the US images with the Allen brain atlas.The temporal trace of each region was derived by averaging the pixel-wise temporal traces within the region.
The SNR of the normalized, visually evoked PA signals was defined as the peak magnitude of the signals divided by the standard deviation of the signals during the 20 seconds before retinal photostimulation.The study included 6 mice.The piezoelectric transducer and CMUT arrays were examined with 3 mice each.T-tests were performed with the null hypothesis that there is no difference between the SNR produced by the two transducer arrays.

Phantom studies with fishing line targets
The US and PAT images of the fishing line were deconvolved with the profiles of the fishing lines.Deconvolving the fishing line images with their actual profiles produces the point spread function, as shown in Fig. 2. Both arrays could accurately locate the fishing lines.With a broader bandwidth, the CMUT array showed 1.3-2.9times better spatial resolution due to its wider frequency bandwidth in Fig. 3.The CMUT array also showed better SNR in Fig. 3, due to the integration of frontend electronics in the imaging probe, as mentioned above.

Photoacoustic imaging of the mouse brain in vivo
Figure 4 shows the US and PAT images acquired by the transducer arrays.As shown in the US images, both systems could capture the anatomies of the inner brain, which assisted in locating the brain regions of interest.The CMUT array appeared to resolve slightly more heterogeneous anatomies in the inner brain.The PAT images acquired by both arrays also showed anatomical details within the inner brain, although not as clearly defined as those in the US images.This is due to the omnidirectional nature of PA emissions [26], which tend to generate more reconstruction artifacts.

Discussion
A limitation of this study is that the piezoelectric transducer array and the CMUT array did not have exactly the same specifications.With this caveat, the side-by-side comparisons indicate that CMUT is comparable to or even better than the piezoelectric array in spatial resolution, penetration, and SNR under our testing conditions.The piezoelectric array shows slightly better lateral resolution, presumably because of its wider aperture.With the advantages of light weight, flexibility, and scalability to 2D arrays, CMUT can be a promising technology for studying brain activities in unanesthetized, unrestrained small animals, which would be challenging if not impossible using piezoelectric transducer arrays.
In Fig. 3, the piezoelectric transducer array shows less consistent resolution vs. depth relationships than those for the CMUT array.This may be due to the lower SNR of the piezoelectric transducer array, which is consistent with the piezoelectric signals' tendency to have greater variability in Fig. 3 rows 1 and 2, and with the SNR comparisons in Fig. 5 right panel.The CMUT and piezoelectric transducer array images in Fig. 4 were taken from two different wild-type mice; we photostimulated each mouse just once to ensure that both mice were fully dark-adapted prior to photostimulation.The hot region at the top of the skull contour in Fig. 4(c) might be caused by a small amount of blood at the skull/membrane interface.Given the US resolution of the PAT modality, the anatomies and response signals inside the brain were not contaminated by the superficial signals.
By imaging the coronal planes of the mouse brain, we have validated the feasibility of using PAT and US images to not only reliably locate the V1 cortex and the subcortical SC but also capture visually evoked hemodynamic responses in these regions.Despite the limited sample size, we were able to show that the CMUT array produces slightly better SNR in imaging the mouse V1, with statistical significance.The Power analysis of this pilot cohort will guide future statistical studies with larger sample sizes.
For 1D transducer arrays, sequential scanning across the brain is required to locate the brain regions of interest.2D matrix arrays provided by CMUT technology will substantially simplify the scanning process by covering the brain volume with single-shot data acquisition.Such a holistic imaging approach will enable rapid, minimally invasive assessments of the mouse visual system, thereby expediting the detection of functional deficits in mouse models of disease.

Conclusions
This is the first demonstration of the feasibility of using PAT-US imaging to capture visually evoked subcortical hemodynamic responses in the mouse brain.We have also shown that the miniaturization process used to make CMUT arrays does not compromise their performance compared to traditional piezoelectric arrays.In future studies, we will aim to fabricate CMUT arrays for brain imaging in anesthetized, free moving mice.

Fig. 1 .
Fig. 1.PAT-US imaging system.(a) Dimensions of the CMUT array.(b) PAT imaging setup.A 3D-printed holder integrated the transducer arrays and the illumination fibers.The illumination was projected into a narrow band at the mouse skull surface.(c) The system schematics.Red lines show the optical paths.Blue lines show the acoustic signal paths.

Fig. 2 .
Fig. 2. Point spread functions produced by imaging a fishing line at 10 mm, 15 mm, and 20 mm depths, respectively.(a) PA images.(b) US images.Scale bars are 3 mm.

Fig. 3 .
Fig. 3. Point spread function profiles in Fig. 2. The distance between the fishing lines and the transducer array surfaces is coded in color.In Rows 1 and 2, red, green and blue curves represent point spread functions at 10 mm, 15 mm and 20 mm depths, respectively.In Row 3, red and blue circles represent CMUT and piezoelectric arrays, respectively.

Fig. 4 .
Fig. 4. Images acquired by the CMUT and piezoelectric arrays.(a) Allen brain atlas showing the plane with V1 and SC regions.(b-c) US images.(d-e) PAT images.(f-g) hemodynamic responses on top of the US images.Scale bars: 3 mm.V1: primary visual cortex.SC: subcortical superior colliculus.

Figure 4 (
Figure 4(f),(g) shows the identified visual regions in the US images and the peak magnitudes of the hemodynamic responses during the monitored 60-sec interval.Figure 5(a),(b) shows the temporal traces within the V1 and SC regions.Figure 5(c) shows statistics of the SNR in the V1 and SC measurements using the two transducer arrays.The SNR produced by the CMUT array in V1 was 5 dB higher than that produced by the piezoelectric transducer array (n = 3 for each group, p = 0.04).The two arrays showed comparable SNRs in the SC.

Figure 5 (
Figure 4(f),(g) shows the identified visual regions in the US images and the peak magnitudes of the hemodynamic responses during the monitored 60-sec interval.Figure 5(a),(b) shows the temporal traces within the V1 and SC regions.Figure 5(c) shows statistics of the SNR in the V1 and SC measurements using the two transducer arrays.The SNR produced by the CMUT array in V1 was 5 dB higher than that produced by the piezoelectric transducer array (n = 3 for each group, p = 0.04).The two arrays showed comparable SNRs in the SC.

Figure 5 (
Figure 4(f),(g) shows the identified visual regions in the US images and the peak magnitudes of the hemodynamic responses during the monitored 60-sec interval.Figure 5(a),(b) shows the temporal traces within the V1 and SC regions.Figure 5(c) shows statistics of the SNR in the V1 and SC measurements using the two transducer arrays.The SNR produced by the CMUT array in V1 was 5 dB higher than that produced by the piezoelectric transducer array (n = 3 for each group, p = 0.04).The two arrays showed comparable SNRs in the SC.

Fig. 5 .
Fig. 5. Hemodynamic responses in visual regions.(a-b) Hemodynamic responses in V1 and SC regions, respectively.(c) Statistics of the SNR of the hemodynamic responses.* indicates a significant difference in the t-test.N = 3 for each group.Piezo.: Piezoelectric transducer.