Research into brain functionality and neurodegenerative disorders critically depends on availability of bioimaging tools capable of non-invasive observations with high resolution and specific contrast [1]. Magnetic resonance imaging (MRI) has enabled a better understanding of brain function, structure, hemodynamics, and connectivity abnormalities in brain diseases, becoming a workhorse in neuroimaging [2]. Mapping of brain activity can also be performed with magnetoencephalography systems. Nuclear imaging technologies, such as positron emission tomography and single-photon emission computed tomography, have further enabled the molecular diagnosis and monitoring of brain diseases. However, these well-established technologies are commonly associated with high installation and maintenance costs, and use of ionizing radiation. These drawbacks have fostered the development of new imaging approaches that can complement or enhance imaging performance and provide better affordability and operability. For example, functional near-infrared spectroscopy capitalizes on optical contrast to measure changes in both oxygenated (HbO₂) and deoxygenated hemoglobin (Hb), which is not possible with functional MRI (fMRI) [3]. However, strong light diffusion in living tissues severely limits the achievable resolution. Functional ultrasound (fUS) provides high sensitivity for imaging cerebral blood flow as has been shown e.g., by monitoring brain activity in newborns through the fontanelles, where the skull is not fully developed. Acoustic distortions induced by the skull are much more significant in adults [4]. Nevertheless, fUS could still be used to reveal brain microvascular structures and hemodynamics through the so-called acoustic windows such as the temporal bone. Ultrasound localization microscopy can further break through the diffraction-limited spatial resolution limits thus revealing microcirculation down to the capillary level [4]. The main limitation of fUS is poor molecular specificity, whilst its sensitivity depends on the direction of blood flow relative to the acoustic axis.

MSOT has been shown to provide unique capabilities for visualizing murine brain activity, including high spatiotemporal resolution and functional specificity [5]. The main advantages of MSOT stem from its optical contrast, which is fundamentally different from the blood flow readouts of fUS. In particular, its five-dimensional (real-time spectroscopic three-dimensional) imaging capability enabled new insights into large-scale neuronal activity and the accompanying hemodynamic changes [6]. Moreover, molecular MSOT imaging of optical probes targeting amyloid-beta deposits and tau fibrils has enabled visualizing specific accumulation of aggregates in animal models, multiparametric characterization of glioblastoma tumors, and neuroinflammation in ischemic stroke [7], which may serve to define new diagnostic biomarkers. MSOT imaging, however, faces similar transcranial imaging challenges as fUS. The relatively thin murine skull bone has been shown to induce acoustic aberrations in broadband optoacoustic (OA) signals, particularly in the high frequency components, leading to observable distortions and loss of contrast and resolution of the images [8]. Much more significant acoustic distortions are induced by the thick human skull, which further supports complex propagation patterns involving longitudinal to shear mode conversion and guided acoustic waves [9,10]. Efforts have been devoted to the development of new algorithms trying to correct for skull-induced aberrations [11–13], for which data from other modalities such as x ray computed tomography and MRI can be used [14,15]. More recently, MSOT has been shown to provide powerful capabilities to map brain activity matching fMRI readings in patients who underwent hemicraniectomy [16]. However, no transcranial in vivo MSOT imaging of human subjects has thus far been achieved.

In this Letter, we demonstrate the basic feasibility of visualizing the cerebral vasculature through the temporal bone in humans with transcranial MSOT imaging. We opted for imaging the temporal cortex as it is affected in many brain diseases while the skull is relatively thin is this region. The results were validated and registered with head-to-head TOF-MRA and T₁-weighted structural MRI. The imaging system consisted of a spherical matrix array (Imasonic SaS, France) accommodating 256 piezocomposite elements and covering an angle of 90° [17]. All the elements have an approximate size of 3 × 3 mm², central frequency of 4 MHz, and detection bandwidth of ∼100%, providing an almost isotropic resolution of 200 µm in a region close to the center of the spherical array geometry. The achievable resolution and overall performance have previously been characterized in detail elsewhere [17]. The excitation light was provided with an optical parametric oscillator laser tunable between 680 and 1250 nm and delivering ∼40 mJ of energy per pulse at a pulse repetition frequency up to 100 Hz (Innolas, GmbH, Germany). The optoacoustically generated signals were digitized at 40 Mega Samples per second with a custom-designed data acquisition system (Falkenstein Mikrosysteme, Germany) triggered with the Q switched output of the laser and transmitted to a personal computer via 1-Gbit/s ethernet. The pulse duration of the laser was less than 10 ns.

Non-invasive MSOT imaging of the brain of a healthy female volunteer (Fitzpatrick scale type-II skin) was performed through the temporal bone with the spherical array operated in a hand-held mode. For this type of skin, the melanin content is relatively low thus the induced light attenuation is insignificant, which may differ for other skin types with more pigmentation. A custom-designed holder 3D-printed in polylactic acid was attached to the array and filled with agar (1.3% agar powder w/v) to guarantee acoustic coupling [18]. The laser beam was guided through a fiber bundle with five output arms [Fig. 1(a)]. The laser fluence at the skin surface was ∼12 mJ/cm², well below safety thresholds for human exposure [19]. To minimize motion artifacts in multi-spectral datasets recorded in a handheld mode, the acquisitions were limited to three wavelengths, which is merely sufficient for quantifying oxygen saturation (sO₂). The laser wavelength was therefore hopped between 700, 800, and 1064 nm at a pulse repetition frequency of 25 Hz. The wavelengths were selected to maximize sensitivity for unmixing the oxygenation readings. The probe was then slowly scanned along the left temporal bone. During the image acquisition, the eyelids were closed and further protected with black tape to avoid any accidental exposure. Image reconstruction was performed with a graphics processing unit implementation of a back-projection algorithm [20]. This reconstruction method enabled real-time rendering to preview the images during acquisition, thus facilitating the localization of the structures of interest. The images were reconstructed under uniform speed of sound assumptions, and manually adjusted to a value that maximized contrast (1495 m/s). This approach is not optimal, but accurate correction for human skull aberrations requires much more sophisticated modeling. Spectral unmixing was performed with a standard least-square fitting approach to the spectra of oxygenated and deoxygenated hemoglobin after normalizing the signals with the wavelength-dependent laser fluence [18] in MATLAB (MathWorks, USA). Individual 3D images were then compounded [21] to provide a larger field-of-view spanning 30 × 30 mm in the lateral dimension.

 

Fig. 1. Transcranial MSOT imaging of human brain through temporal bone using a hand-held probe. (a) Illustration of the multifiber hand-held MSOT probe (diameter 6 cm) and (b) imaging through temporal bone. The laser beam was guided through a fiber bundle with five output arms to facilitate uniform illumination pattern on the skin surface (beam diameter at the tissue surface approx. 15 mm). (c)–(e) MSOT images acquired at 700, 800, and 1064 nm. Field-of-view 10 × 10 × 25 mm. (f)–(h) Spectrally unmixed distributions of HbO₂, Hb, and the resulting sO₂ of the deep vessel inside the brain (indicated by green box in panel c). (i) MSOT image recorded at 800 nm (MIPs shown in the axial, sagittal, and coronal orientations). m, medial; l, lateral; a, anterior; p, posterior; d, dorsal; v, ventral. Scale bar = 20 mm in panels (a) and (b), 3 mm in panels (c) and (d), and 1 mm in panels (f)–(i).

Download Full Size | PDF

In-human transcranial brain images acquired with MSOT through the temporal bone are shown in Fig. 1(b). The effective imaging depth was sufficient to clearly visualize the temporal cortex and superior middle cerebral vein (SMCV), also known as the Sylvian vein, located at an approximate depth of 13 mm from the skin surface [lower parts of Figs. 2(a)–2(c)]. This vessel was identified as a vein based on a sO₂ value in the 60–70% range [Figs. 1(f)–1(h)], whilst sO₂∼100% is usually expected in arteries. Similar values have previously been reported with near-infrared spectroscopy [22], QSM-MRI [23], and are also consistent with previous MSOT measurements in the murine brain and other regions of the human body. Additionally, the vessel observed in the MSOT image matches the SMCV in the MRI image. Note that the MSOT-derived sO₂ measurements are affected by the so-called spectral coloring effects (wavelength-dependent light attenuation). However, reliable differentiation between arteries and veins has been reported at depths of up to 1–1.5 cm [24]. The deep vessel SMCV was more prominent at 800 nm and 1064 nm, arguably due to the higher light attenuation at shorter wavelengths. The superior temporal vein (STV) could also be localized [upper parts in Figs. 1(c)–1(e)], with more prominent contrast manifested at 700 nm. The wavelength-dependent signal patterns in deep tissue differed from the more superficial signals generated in the skin/muscle, indicating that the vascular patterns in the brain were not associated with skull reflections.

 

Fig. 2. Registration of MSOT and MRA of human brain. (a) MSOT image recorded at 800 nm. MIPs are shown in the axial, coronal, and sagittal orientations of the temporal cortex region; (b) TOF-MRA MIPs are shown in axial, coronal, and sagittal view for illustration purposes; and (c) overlay of panels (a) and (b) in axial, coronal, and sagittal view. (d) Zoomed-in of MSOT image recorded at 800 nm and overlay of MSOT image recorded at 800 nm overlay with TOF-MRA in axial, coronal, and sagittal view; scale bar = 10 mm

Download Full Size | PDF

To map the exact thickness of the temporal bone and location of the STV and deeper cerebral vessels, TOF-MRA and T₁- weighted MRI scans were performed in the same subject at University Hospital Zurich using a GE Signa Premier 3T scanner (Fig. 2). Standard clinical 3D TOF MRA sequences (slice thickness = 1.2 mm, repetition time (TR) = 21 ms, time to echo (TE) = 3.4 ms, flip angle = 20 degrees, autocalibrating reconstruction for Cartesian imaging) with HyperSense acquisition acceleration and a 48-channel head coil provided optimal vascular contrast. A native T₁-weighted MR magnetization-prepared 180 degrees radio frequency pulse and rapid gradient-echo (MP-RAGE) were used as structural reference (slice thickness = 1 mm, interslice gap = 0.5 mm, image matrix 512 × 512 × 312, voxel size of 0.4688 × 0.4688 × 0.5 mm, TR = 2191.36 ms, TE = 2.996 ms, inversion recovery = 900 ms, flip angle = 8 degrees) [25]. PMOD software (PMOD Technologies, Switzerland) was subsequently used for segmentation and registration of MRI and MSOT data.

Owing to their very different FOV scales, registration using deformable or rigid algorism between MRI and MSOT images was challenging. Additionally, MRI and MSOT acquisitions cannot be performed simultaneously due to incompatibility of the two current scanners. Thus, manual MSOT-MRI registration assisted with a fiducial marker was used in this study: identification of the region of interest (temporal window) for MSOT imaging was achieved by drawing a cross on the skin close to the temporal bone of the participant. This also served as a fiducial marker to facilitate the registration process. In addition to the fiducial marker, co-registration of the MSOT and MRA data (Fig. 1) was further validated by locating the SMCV in the temporal cortex in the MSOT images. The SMCV and STV vessels measured approximately 1 mm in diameter, as seen from the TOF-MRI (Fig. 2, Visualization 1). Similar STV dimensions were rendered with both modalities, although the SMCV appears smeared in the MSOT images (Fig. 1). This is arguably attributed to severe acoustic aberrations introduced by the skull. The skull thickness was estimated to be approximately 1.3–3 mm in the temporal bone area with the T₁-weighted images, which is within the range of previously published results [26]. To assess the effects of acoustic aberrations in the MSOT images, we imaged a 200-µm microsphere through different regions of an ex vivo skull sample corresponding to different thicknesses. It is shown that, despite some distortions, the microsphere can be imaged through 1.5 mm of the skull bone (Fig. 3), corresponding approximately to the skull thickness in the temporal window.

 

Fig. 3. Skull influence on MSOT resolution and signal attenuation. (a) MSOT images of a 200-µm microsphere through different regions of a human skull of different thicknesses. Scale bar = 2 mm. (b) Achievable resolution as a function of skull thickness estimated as the full width at half maximum (FWHM) of the MSOT image. (c) Signal attenuation as a function of skull thickness.

Download Full Size | PDF

This work demonstrates the basic feasibility of performing non-invasive MSOT brain angiography through the temporal bone in humans. Our results indicate that human skull induces strong acoustic aberrations resulting in significant distortion of deep vascular structures (Fig. 3). We further investigated to what extent the images were afflicted with other potential sources of image distortions. In particular, pressure waves generated within subcutaneous blood vessels undergoing acoustic reflections in the skull may result in spurious structures mirrored into the brain. However, in our experiments, no strong contrast was generated at the skin surface that could potentially result in strong reflections. Furthermore, the vascular structures in the MSOT images acquired at different wavelengths exhibited a distinctive absorption spectrum. Finally, TOF-MRA images also served to validate the actual location of the vessels identified in the MSOT images. We are thus confident that the observed structures actually correspond to the pial vasculature.

MSOT imaging is based on tissue excitation with non-ionizing NIR radiation that is safe for human use. The laser energy density employed in this work was below safety standards causing no thermal or other adverse effects. Several experimental aspects can be further optimized. For example, the efficiency of excitation light delivery can be improved based on a photon recycler that has been shown to recover ∼50% of the optical energy loss [27]. US attenuation is generally known to increase with frequency [28], e.g., resulting in ∼80% signal loss at 1 MHz through the full skull thickness [29], which makes it easier for the low frequency signal components to traverse the skull with attenuation and aberrations. MSOT at lower frequencies, e.g., at 1 MHz, can still render sufficient resolution to provide valuable information. Indeed, the acquired signals generated within the brain have a dominant low-frequency component due to the significant increase in acoustic attenuation with frequency. The achievable depth in MSOT imaging is generally limited to approximately 2–3 cm due to light attenuation. For transcranial imaging, acoustic attenuation is the dominant factor affecting the effective penetration depth. Therefore, for regions where MSOT imaging is not impeded by the skull, imaging at depths of 1 cm within the brain is expected to be feasible. MSOT imaging of the brain is facilitated in regions with a thin skull, namely, the acoustic windows. The diminished thickness of the temporal bone facilitates US transmission, which has previously been exploited for pulse-echo US imaging of anterior, middle, and posterior cerebral artery [30]. Apart from the temporal window, other transcranial windows, such as the submandibular window, the suboccipital window, or the transorbital window, have been identified as suitable for Doppler US imaging [30]. Likewise, acoustic sensors may be positioned within the nasal cavity or ocular regions, which have been shown to facilitate US access to the brain [14]. The transorbital window may not be usable for MSOT imaging, as it involves light propagation through the eyes [31].

Apart from causing strong acoustic dispersion and attenuation, the high acoustic impedance of the skull bone (∼7.7 × 10⁶ Rayls [32]) relative to the soft brain tissue results in loss of spatial resolution when assuming a uniform speed of sound for the MSOT reconstruction procedure. Accurate modeling of transcranial US propagation is not straightforward, as it implies considering mode conversion at surfaces and guided wave propagation [18]. The development of reconstruction algorithms accounting for these effects could potentially be facilitated with accurate prior anatomical information available in the MRI images. Indeed, the feasibility of hybridization between preclinical MSOT and MRI has recently been demonstrated [33]. A method for correcting the distortion in pulse-echo transcranial US images acquired through the temporal bone has recently been proposed based on estimating aberration delays from the backscattered US waves in individual microbubbles [10]. Combining MSOT and pulse-echo US may then further enable correcting for acoustic distortions in the MSOT images. Additionally, absorbing microparticles could also be localized and tracked individually in real time with MSOT [34]. These and similar particle localization approaches may also facilitate the development of dedicated MSOT reconstruction algorithms tailored for transcranial imaging, e.g., based on the optoacoustic memory effect.

The MSOT capacity for transcranial imaging in humans can potentially define new application niches in clinical brain imaging. The recent validation of MSOT readings in humans with clinical fMRI anticipates the great potential of MSOT to study brain function in health and disease. MSOT has recently demonstrated high molecular sensitivity for targeting various brain diseases in preclinical models and can potentially be used for defining new biomarkers for in vivo clinical brain research and diagnosis.