Authors: Emily Rathbone, Dan Fu
Categories: Article
Source: The journal of physical chemistry. B
Authors: Emily Rathbone, Dan Fu
The intimate relationship between neuronal activity and cerebral oxygenation underpins fundamental brain functions like cognition, sensation, and motor control. Optical imaging offers a non-invasive approach to assess brain oxygenation and often serves as an indirect proxy for neuronal activity. However, deciphering neurovascular coupling—the intricate interplay between neuronal activity, blood flow, and oxygen delivery—necessitates independent, high spatial resolution, and high temporal resolution measurements of both microvasculature oxygenation and neuronal activation. This Perspective examines the established optical techniques employed for brain oxygen imaging, specifically functional near-infrared spectroscopy, photoacoustic imaging, optical coherence tomography, and two-photon phosphorescent lifetime microscopy, highlighting their fundamental principles, strengths, and limitations. Several other emerging optical techniques are also introduced. Finally, we discuss key technological challenges and future directions for quantitative optical oxygen imaging, paving the way for a deeper understanding of oxygen metabolism in the brain.
The human brain accounts for only 2% of the body’s weight but consumes nearly 20% of the body’s chemical energy in the form of oxidative phosphorylation.^1^ Most of this energy is used to maintain the resting membrane potential of neurons and to support neurotransmission, processes that are crucial for normal brain function, including cognition, sensation, and motor control.^2^ Given the limited energy reserves within the brain, continuous oxygen and glucose supply through blood flow is essential to sustain oxidative phosphorylation and, consequently, brain function. Disruptions in oxygen supply can lead to dire consequences, ranging from subtle cognitive impairments to severe neurological damage, as seen in stroke^3^, trauma^4^, and neurodegenerative diseases.^5^
Oxygen is carried by the hemoglobin protein (Hb) in red blood cells (RBCs). The brain’s complex vasculature plays a pivotal role in delivering oxygen-carrying RBCs to this metabolically demanding organ, ensuring its proper function and survival. Comprised of a dense network of arteries, capillaries, and veins, the cerebral vascular system is intricately designed to meet the brain’s high energy needs.^6,7^ Oxygen-rich blood reaches the brain through the internal carotid and vertebral arteries, which branch out to different brain areas. The cortical surface features the highly complex and redundant pial arteriolar mesh. From this mesh, penetrating arterioles extend into the brain tissue which bifurcate extensively to form a finely-tuned capillary network, facilitating efficient oxygen and nutrient exchange at the cellular level.^8^ The neurovascular unit, a functional ensemble of endothelial cells, neurons, and glial cells, orchestrates the coupling of cerebral blood flow with neuronal activity, a phenomenon also known as neurovascular coupling.^9^ This mechanism ensures that regions of the brain with heightened activity receive increased oxygen and nutrient supply to meet local metabolic demands. However, how this regulation is achieved through the coordination of arteriole, capillary, and glia cells remains elusive. It is unclear how oxygen is delivered and distributed at the microvascular level, particularly in the brain’s capillary beds where the actual exchange of gases occurs.
Given the pivotal role oxygen plays in brain function, the measurement of oxygen in the brain is a cornerstone of understanding neurovascular coupling and cerebral physiology and pathophysiology. Imaging plays a central role in oxygen measurement in the brain due to the heterogeneous oxygen distribution in the vasculature and brain parenchyma. The challenge is to visualize the complex vascular network and dynamics of oxygenation changes within the full depth of the parenchyma with sufficient spatial and temporal resolution. This challenge has driven the development of optical imaging technologies for decades. This Perspective reviews these technologies, discussing their strengths, limitations, and future prospects to overcome current challenges in the field.
Many techniques have been developed for oxygen measurement in the brain and other tissues. Functional magnetic resonance imaging (fMRI) is widely used to monitor brain activity through blood oxygen-level-dependent (BOLD) contrast changes, which is sensitive to the local decrease in blood oxygenation due to the increased metabolic activity at activated brain regions. Due to the magnetic properties of blood, BOLD contrast is sensitive only to the change in deoxygenated hemoglobin concentration and thus provides limited information about the oxygen content.^10^ Positron emission tomography (PET) measures cerebral oxygen use and blood flow using an ^15^O radiotracer but oxygen levels are not directly measured with this technique. Additionally, PET involves potentially harmful ionizing radiation.^11^ Pulse oximetry, another method, distinguishes oxygenated from deoxygenated hemoglobin by light absorption differences at two wavelengths, providing continuous transcutaneous blood oxygen saturation (sO2, the concentration ratio of oxygenated Hb to total Hb) measurements.^12^ However, pulse oximetry only measures sO2 of arterial blood because it uses the pulsatile nature of the cardiac cycle to extract and normalize the cyclic signal, and only arterial blood volume changes during the cardiac cycle.^13^ Moreover, it only measures the average sO2 of tissue traversed by the emitted light at a single location. Nevertheless, this basic principle of using the difference in absorption between oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (HbR) to measure sO2 underlies many of the more advanced quantitative optical imaging techniques as seen in the following sections.
While fMRI, PET, and pulse oximetry are essential in both preclinical and clinical settings, they lack the spatial resolution to detail oxygen metabolism and neurovascular coupling at the microvascular level. Optical imaging, in contrast, offers superior resolution and faster measurements, potentially overcoming these limitations. Figure 1 summarizes the current and potential optical imaging techniques available for oxygen measurements in the brain. They can be roughly divided into two those based on HbO2 and HbR absorption difference and those relying on oxygen-dependent properties. In the following sections, we will describe each technique and discuss its strengths, weaknesses, and potential applications.
It is worth noting that some techniques are appropriate for humans while others are only applicable to animal models, particularly mice or rats. Humans have thicker skulls and gray matter than rodents. Invasive measurements (e.g., craniotomies) are much less common in humans. There are also functional differences between human and rodent brains. For example, the cerebral metabolic rate of oxygen in gray matter is around 3.3–5.5 ml/100 g/min in humans and is closer to 6.4 ml/100 g/min in mice.^14,15^ Recognizing differences in oxygen demand and consumption in animal models is important when applying findings from animal models to human neuroscience studies. The Perspective focuses on the technical aspect of optical imaging instead of applications. However, when appropriate, the application scope of each technique will be discussed.
Functional near-infrared spectroscopy (fNIRS) builds on pulse oximetry principles, using NIR light’s absorption differences between HbR and HbO2 (Figure 2a) to determine concentration changes of both in the probed cerebral volume.^16–18^ The NIR light, highly scattered upon penetration, creates a diffuse light field that illuminates the cortex (Figure 2b). Light sources, placed on the scalp or skull, work alongside detectors (avalanche photodiodes or photomultiplier tubes) to measure light absorption. The initial demonstration of fNIRS dates back to 1977, when Jöbsis first established the feasibility of fNIRS in monitoring the cortical hemoglobin concentration changes in vivo.^19^ The basic principle of fNIRS builds on the modified Beer-Lambert Law, which adds a scattering-dependent loss parameter G.
(1)OD=logI0I=∑iεiλcilDPFλ+Gλ
where OD represents the optical density, I0 is emitted light intensity from the source, I is detected light intensity on the detector, i denotes the chromophores (HbO2 and HbR), ε is the absorption coefficient, c is the concentration, l is the distance between the source and detector, and DPF represents the differential path length factor (accounting for the increased distance light travels compared to l due to scattering). When G is assumed to be a constant during the measurement, the change in OD is only due to concentration changes of HbO2 and HbR:
(2)ΔOD=∑iεiλΔcilDPFλ
By measuring OD changes at two or more wavelengths, fNIRS can calculate changes in concentrations of HbO2 and HbR, assuming uniform concentration shifts within the observed area. This assumption may reduce accuracy for localized sO2 changes.^20^ With multiple light sources and detectors, measurements of many tissue locations (or channels) can be acquired. This relatively simple implementation of fNIRS, known as continuous wave fNIRS (CW-fNIRS), is widely used due to its simplicity, low cost, scalable channels, and high temporal resolution.^14^ However, it can only determine relative Hb concentration changes due to the unknown scattering-dependent loss G and differential path length factor DPF.
The limitations of CW-fNIRS can be addressed by collecting additional information from the detected light to separate absorption from scattering. Time-domain fNIRS (TD-fNIRS) uses pulsed light sources (typically 10–100 ps) and measures the photon time-of-flight distribution (Figure 2c).^21^ Photon migration modeling using the diffusion equation or Monte Carlo simulation is needed to determine the absorption and reduced scattering coefficients (μa and μs’, respectively), which are then used to calculate sO2 and total Hb concentration.^22^ Frequency domain fNIRS (FD-fNIRS) is another technique that can be used to determine both sO2 and absolute hemoglobin concentration by separating the absorption and scattering effects using modulated NIR light at high frequencies and measuring both the intensity attenuation and phase shift. The mathematical principle, instrumentation, and data analysis of time-domain and frequency-domain fNIRS are beyond the scope of this Perspective. We refer interested readers to published reviews for details.^22,23^
Regardless of which source and detection technique is used, fNIRS modalities share many common features and applications. For brain imaging, it is desirable to have many source-detector channels to spatially resolve hemodynamic features across different brain regions. Multi-channel fNIRS enables 3D imaging through image reconstruction, known as diffuse optical tomography (DOT), capturing local blood volume and sO2 shifts.^24^ DOT involves solving the “forward problem” of light distribution prediction and the “inverse problem” of reconstructing internal absorption and scattering based on overlapping multi-channel measurements.^25^ However, the inverse problem is ill-posed and underdetermined, making it challenging to define the solution.^17^ Additionally, DOT suffers from low signal-to-noise (SNR) ratio, low temporal resolution (arising from the need for signal averaging), and noise concerns due to poor optode (source and detector unit) contact and movement.^26,27^ Intense efforts have focused on addressing these limitations to improve DOT’s usefulness in clinical imaging of the brain.^28^ For example, Cai et al. recently adapted a nonlinear source localization method to solve the inverse problem of DOT reconstruction and achieved robust results in low SNR conditions and accurate DOT reconstructions compared to fMRI activation mapping.^29^ Improvement in the number of channels using a higher density array of optodes can further push the resolution limit with image quality approaching that of fMRI or even better.^30^ Nevertheless, because fNIRS uses diffuse light for measurements, they suffer from intrinsically poor resolution, typically ranging from a few millimeters to centimeters. Recently, high density DOT has pushed the resolution down to an impressive 1 mm^3^ at a depth of 7 mm.^31^ However, even at mm resolution, it is unable to resolve single arterioles or venules, which are 1–2 orders of magnitude smaller.
The main advantages of fNIRS over other optical imaging techniques are deep tissue penetration, portability, and applicability to humans. Leveraging the NIR range’s lower light absorption and scattering, fNIRS achieves several centimeter imaging depths that readily probe through the scalp or skull, allowing non-invasive monitoring of cerebral oxygenation and hemodynamics in both healthy individuals and patients with neurological conditions such as stroke, traumatic brain injury, and neurodegenerative diseases.^24,32–35^ fNIRS has advanced from relatively simple measures of blood-oxygen changes in its early days to a sophisticated method of recording real-time brain responses associated with a wide variety of activities and cognitive tasks.^15,36^ Unlike fMRI, the portability of fNIRS enables subjects to move, speak, and interact naturally during studies, gaining functional information that was previously inaccessible.^37^ fNIRS also provides a suitable platform for imaging brain activity in awake infants, which is especially challenging with traditional methods due to the need for sedation, bodily restraint, and contrast agents. For example, Frijia et al. developed the first high-density DOT device for infants and compared hemodynamic responses during audio-video stimuli with increased spatial localization and high SNR, permitting direct association to the cortical anatomy (Figure 2d).^38^
An fNIRS-related optical imaging technique is diffuse reflectance imaging of brain hemodynamics. It uses a camera to capture wide-field reflected images at two or more visible wavelengths.^39^ Similar to fNIRS, the modified Beer-Lambert law can be used to extract relative concentration changes of HbO2 and HbR. This method offers much-improved lateral resolution at the expense of imaging depth. Only surface vessels of the exposed cortex in animal models can be imaged to detect changes in hemodynamics at high speed.^39^ Its resolution is not on the capillary level, but it can be readily used in awake animals with high enough temporal resolution for studies such as mapping stimulation response.^40^
Photoacoustic imaging (PAI) is another powerful imaging technique that is widely used in mapping oxygen metabolism in the brain. It indirectly measures chromophore absorption through ultrasound detection. When short laser pulses (typically nanoseconds) are absorbed by molecules in tissue, the resulting rapid rise in temperature causes thermoelastic expansion and contraction and generates acoustic waves, which an ultrasound transducer can detect (Figure 3a).^41,42^ Therefore, PAI combines the high contrast of optical absorption with the deep penetration of ultrasound detection because acoustic waves have three orders of magnitude weaker scattering than light.^43^
The principle and theory of photoacoustic imaging has been extensively reviewed in previous publications.^44–46^ Here we only focus on sO2 imaging in the brain. PAI can specifically image HbO2 and HbR by tuning the wavelengths to the absorption peaks of each, typically in the visible region owing to their strong and distinctive absorption peaks.^45^ Functional imaging is achieved by analyzing the photoacoustic signals for absorption at two wavelengths^46^:
(3)Pλi,x,y=ϕλi[εHbRλicHbRx,y+εHbO2λicHbO2x,y
where φ(λi) is the local optical fluence at wavelength λi, ɛ is the molar absorptivity, c is the concentration of the indicated species, and P is the reconstructed photoacoustic image. Assuming optical fluence linearly correlates with incident light intensity, sO2 levels can be calculated using two wavelengths. This straightforward approach is effective for shallow depths but faces accuracy issues for deeper tissues due to tissue optical properties, light delivery, and wavelength impact.^47^ To estimate local fluence variations, more complex models using either mathematical or orthogonal imaging methods were developed.^46^ DOT measurements can best estimate local fluence and reduce fluence-related errors in sO2 measurements^48^; however, the low resolution of DOT limits its use in high-resolution photoacoustic imaging. Machine learning-based approaches have also shown promise and are faster than model-based algorithms, which is beneficial for real-time hemodynamic imaging.^49^ Another possible source of error in photoacoustic imaging of sO2 is the presence of additional chromophores in tissue such as cytochrome c.^50^ The impact of these absorbers on the sO2 accuracy awaits further investigation.
PAI is categorized into photoacoustic microscopy (PAM) and photoacoustic computed tomography (PACT), depending on whether the illumination is focused or widefield. PAM offers high resolution through focused illumination and acoustic detection. Imaging is achieved by scanning the sample or the laser beam in two dimensions across the field of view (FOV). Depth-resolved signal from a single illumination point on the sample gives rise to an A-line. Scanning the illumination along multiple locations forms a 2D cross-sectional image called a B-scan. In comparison, an en face image is defined as the image perpendicular to the depth axis. One distinctive advantage of PAM compared to other optical imaging techniques is that its resolution and imaging depth can be scaled to suit a specific application.^44^ PAM has two variants. Optical resolution PAM (OR-PAM) uses a tightly focused laser with a high numerical aperture (NA) microscopic objective and the resolution is determined by the optical focus 0.51λ/NA.^44,51^ In contrast, acoustic resolution PAM (AR-PAM) uses a loosely focused laser beam and the resolution is determined by the acoustic focus 0.71λa/NA~a. Here λa~ is the acoustic wavelength (λa = va/fa, where va is the speed of sound in the medium, and fa is the central frequency) and NA~a~ is the NA of the focused ultrasonic transducer. In both cases, the laser excitation and acoustic detection are confocally aligned to maximize sensitivity. Due to light scattering, OR-PAM can only image ~1 mm deep (ballistic or quasi-ballistic regime), but the lateral resolution can reach a few hundred nanometers. AR-PAM is typically used in the optically diffusive regime (>1 mm). The imaging depth is primarily determined by the frequency-dependent acoustic attenuation. Typical AR-PAM systems employ a transducer with a central frequency greater than 20–50 MHz to provide a sub-100 μm lateral resolution with an imaging depth of a few millimeters, sufficient to resolve major cortical blood vessels transcranially in live mice. To reach an imaging depth >1 cm, a low-frequency (<10 MHz) transducer can be used at the expense of the spatial resolution. It has been shown that the maximum imaging depth of PAM is approximately proportional to its lateral resolution.^44^
For both AR-PAM and OR-PAM, it is important to consider the axial resolution and depth of focus besides lateral resolution and imaging depth. Signals along the axial dimension (A-line) are derived from the time-resolved acoustic detection. Axial resolution is determined by the bandwidth (Δf) of the ultrasonic transducer, calculated as 0.88c/Δf, where c is the speed of light.^44^ Typical resolutions are in the range of 15–50 μm. While high-frequency transducers can be used to improve axial resolution, they suffer from significant acoustic loss and low penetration depth.^52^ The other aspect that is less discussed is the depth of focus. With either optical focusing or acoustic focusing, optimal lateral resolution can only be achieved within the range of the depth of focus around the focal zone, which is much lower than the maximum imaging depth. Refocusing can be used to improve resolution, but 3D image acquisition will take much longer. For OR-PAM, the optical beam can be shaped to have a much larger depth of focus than the traditional Gaussian beam. This can be achieved with a variety of methods, such as dynamic focusing^53^, Bessel beams^54^, and structured illumination.^55^ Cao et al. recently demonstrated needle beam PAM with an extended depth of focus that is 14-fold larger in visible PAM imaging of mouse brain, reaching 2.3 μm lateral resolution over the range of 1 mm.^56^ AR-PAM can also benefit from extended depth-of-focus approaches such as multifocal point transducer^57^, synthetic aperture focusing^58^, and deep learning.^59^
OR-PAM has reached single-RBC resolution and detected its oxygen release along with hemoglobin concentration and flow rate in real-time at 20 Hz B-scan rate with an imaging depth of 300 μm.^60^ This was accomplished by using two visible lasers at 532 nm and 560 nm to sequentially excite the same flowing single cell and acquire two depth-resolved photoacoustic signals to determine HbO2~ and HbR concentration. However, accurate in vivo measurement of absolute sO2 remains a challenge due to wavelength-dependent light attenuation. Yao et al. developed a single wavelength-based OR-PAM system for high-resolution, high-speed transcranial imaging of the mouse brain (Figure 3b).^61^ When excited by a picosecond pulse followed by a nanosecond pulse at 532 nm, HbO2 and HbR display different saturation levels and give rise to different photoacoustic signals, allowing quantification of sO2. They achieved a ~3 μm lateral resolution and 15 μm axial resolution with a 2D imaging rate of 400 Hz over a 3 mm range. This method does not suffer from wavelength-dependent attenuation. However, the imaging depth is limited by optical attenuation at the point where saturation becomes insufficient, which is around 0.7 mm. Whole-brain mouse imaging with microvessel resolution has been achieved with AR-PAM, offering a large FOV volume of 11 × 7.5 × 1.5 mm^3^, 2 Hz 3D imaging rate, and 10 μm spatial resolution (Figure 3c).^62^ The authors noted that their depth of focus was 0.42 mm deep and when imaging at 1 mm, the lateral resolution decreased to ~127 μm due to beam divergence and optical scattering. Applying this high-speed large-field functional PAM technology, they were able to image the entire mouse cortex’s fast morphological and functional changes in response to systemic hypoxia, sodium nitroprusside, and stroke.
An alternative to the scanning-based PAM is widefield illumination, utilized in PACT for rapid brain imaging.^41^ PACT employs transducer arrays to capture diffusive signal from widefield pulsed illumination and reconstructs 3D images. This process involves complex time-domain inverse problem-solving, with additional complications from aberration and attenuation of the photoacoustic wave.^63^ PACT is suitable for deep tissue imaging in larger animals or humans. Low ultrasound frequency (<10 MHz) is used to accommodate large penetration depth (>1 cm). PACT has advanced to measure cortical activation in humans, paralleling fMRI’s capabilities. In a study by Na et al., 1,024 transducers were arranged around the heads of human participants to achieve a 1 cm depth from the cortical surface with a 10 cm diameter FOV, 350 μm spatial resolution, and 2 s temporal resolution, with a SNR of 50.^64^ This setup allowed for precise mapping of language processing areas, showing strong agreement with fMRI in terms of activation area and location (Figure 3d). Along with better sensitivity and specificity, PACT functional brain imaging is cheaper than fMRI machines, and can be used in patients with ferromagnetic implants, claustrophobia, or who have noise sensitivity.^64^ However, PACT’s imaging depth is restricted to a few centimeters due to optical scattering and absorption, and spatial resolution diminishes with depth. Transcranial PACT may be possible with skull-induced aberration correction and sensitivity enhancements to reliably capture subtle blood flow and oxygenation changes.^65,66^
Optical Coherence Tomography (OCT) is an advanced imaging technique that offers high-resolution, cross-sectional visualizations of biological tissues, including the intricate structures of the eye, skin, and brain.^67–69^ It employs a low-coherence light source with a interferometer that combines backscattered light from the sample arm with the reference arm reflection to generate interference patterns (Figure 4a).^70^ Depth-resolved signal (i.e. A-line) is obtained via Fourier transform of the interference in either the time domain or spectral domain. Similar to PAM, by scanning the A-line in one or two dimensions, high resolution (1–10 μm) cross-sectional OCT images (B-scans) or volumes can be generated, respectively. Since its inception in the early 1990s, OCT has become a widely used imaging technique in ophthalmology, with rapid expansion into cardiology, dermatology, gastroenterology, and oncology. There are a plethora of reviews that have been dedicated to the various technological developments and applications of OCT so we refer readers to recent ones for more information.^71^
OCT has been widely used to visualize vasculature and quantify blood flow, especially non-invasively in the eye. This can be achieved with either Doppler OCT^72^, where the phase change of OCT signals from flowing RBCs is measured, or OCT angiography (OCTA)^73^, which uses the phase and/or magnitude differences in consecutive OCT scans to detect blood flow. However, measuring sO2 proves more challenging. To functionalize OCT for measuring sO2, it is necessary to extract the spectroscopic signature from A-line signals due to hemoglobin absorption.^74^ While it is possible to do so in the NIR region^75^, where most OCT systems operate, the use of visible light offers a distinct advantage of stronger absorption with pronounced spectral features, which provides improved contrasts and quantification capability at the expense of increased scattering and lower penetration depth. Using a broadband visible light source, Yi et al. demonstrated that sO2 can be measured in vivo in retinal vessels through spectral analysis of OCT signals.^76^
In visible light OCT (vis-OCT), the spectral information is first collected by either taking the Fourier transforms of windowed spectral regions or by taking the short-time Fourier transform of the complex signal at different depths to determine the OCT spectra I(λi,z) from the vessel wall (Figure 4b).^76,77^ Least squares fitting is then used to fit the log-normalized spectra OD(λi,z) from the posterior vessel at depth z based on the modified Beer-Lambert law to determine sO2 using the following ^78^
(4)ODλi,z=logIλi,zI0λi,z=−2dcHbRεHbRλi+cHbO2εHbO2λi−αIogλi+logAR0
where d is the vessel size, R0 is the reference arm reflectance, and the scattering spectrum of the vessel wall r(λ) was modeled as a power law Aλ^–α^ under the first order Born approximation. The factor of 2 comes from the double pass geometry through the vessel. Equation (4) resembles both equations (2) and (3) in that total absorption from hemoglobin is measured at two or more wavelengths and the ratio between HbO2 and HbR is then used to determine sO2. However, in equation (4), the wavelength-dependent scattering term αlog(λi) contributes significantly to the signal, especially for small vessels. Spectral fitting in regions sensitive to sO2 (520–580 nm) is necessary to calculate sO2~. Shu et al. wrote an excellent review in 2017 that we recommend for further breakdown of the math involved in vis-OCT sO2 measurements.^79^ Many retina imaging applications have been published using this technique. Among these, vis-OCT imaging of microvascular sO2 of larger vessels was achieved for the first time in human retina in 2020.^80^ The authors further improved their results by combining conventional NIR OCT angiography with vis-OCT for sO2 quantification in vessels 20–30 μm in diameter.^81^
One challenge in vis-OCT is the accurate localization of vessels for sO2 quantification. Pi et al. developed an automatic way to detect the posterior boundaries from the highest SNR position within the A-line scans and achieved accurate results in their measurements of major vessels (Figure 4c).^82^ Other considerations in obtaining accurate spectra include system-dependent (e.g. spectrally-dependent roll-off, background bias, and chromatic aberration) and sample-dependent (e.g. specular reflection and heterogenous sample scattering) spectral contaminants that can degrade sO2 quantification accuracy and reduce repeatability.^83^ Liu et al. addressed these challenges by training neural networks to relate spectral measurements to sO2 and provide uncertainty quantification to robustly and reliably quantify the sO2.^84^ Rubinoff et al. developed “adaptive spectroscopic vis-OCT” to adaptively remove spectral contaminants arising from the sample and the system on a vessel-by-vessel basis.^83^ Their results were highly repeatable and accurately determined sO2 values in larger arterioles. In combination with OCT angiography, Pi et al. achieved quantitative sO2 imaging in retinal capillaries in the three retinal vascular plexuses in rats (Figure 4d).^78^ Such capability can potentially provide crucial insights into retinal metabolism and the role of sO2 in retinal pathology.
OCT-based oxygen measurement is also achievable in rodent brains, along with cerebral blood flow quantification, which allows measurement of the cerebral metabolic rate of oxygen.^85^ Compared to retinal oxygen imaging, reports of cortical measurements are fairly sparse^77,86–89^, probably because the imaging depth is limited to 200–400 μm due to strong light scattering.^90^ This was extended to achieve vis-OCT imaging of the entire cortex down to 1000 μm by surgically implanting a microprism (Figure 4e), but this raises concerns about permanent changes to brain morphology, function, and connectivity, even after the mouse is fully healed.^90^ Despite the rapid advance in vis-OCT techniques and applications, there are some fundamental drawbacks for brain applications. Even in the eye, vis-OCT is limited by visible absorption of the retinal pigment epithelium and the choroidal pigment, making imaging beyond the superficial retinal layers challenging.^91^ Vis-OCT typically uses a high relative intensity noise (RIN) supercontinuum light source, and when combined with low powers and stronger attenuation, low SNR is a common issue, although Wang et al. recently achieved shot-noise limited imaging in humans with the development of a dual-channel system.^91^ Brain imaging applications have not been extensively explored yet and it remains unclear how accurate the sO2 measurement is in the brain, especially at the capillary level, due to the much stronger scattering compared to the eye and lower SNR in capillaries.
Besides using intrinsic properties of Hb for oxygen measurement, exogenous dyes that are sensitive to oxygen can be used to quantify oxygen partial pressure (pO2) directly in animal models. These dyes, upon excitation, transfer energy to oxygen’s unpaired electrons, a process modulated by local oxygen levels, known as phosphorescence quenching.^92^ Higher pO2 enhances quenching, diminishing the phosphorescence intensity and lifetime. Due to their prolonged lifetimes, phosphorescent probes are highly oxygen-sensitive.^93^ The lifetime can be determined by continuously collecting photons emitted at different times and then fitting the resulting decay curve with an exponential function. The pO2 is then estimated from the lifetime with an empirical equation after calibration.
One of the most widely used probes for in vivo brain oxygen imaging is PtP-C343, a Pt tetraarylporphyrin probe developed by Sergei Vinogradov’s group in 2008.^94^ It features a two-photon excitable antenna yielding high quantum efficiency and effective energy transfer to its Pt porphyrin core. The lifetime of phosphorescence is solely a function of pO2. This probe has been widely adopted for microvessel oxygenation studies in rodent brains through two-photon phosphorescence lifetime microscopy (2PLM).^95–99^ A newer probe, “Oxyphor 2P,” reported by the same team in 2019, promises even faster and deeper tissue (~600 μm) oxygen mapping (Figure 5a–c).^93,100^
Compared to the label-free methods based on hemoglobin absorption, 2PLM of Oxyphor 2P and related oxygen probes offer distinct they can measure both vascular pO2 and tissue pO2 with intravascular injection and local tissue injection of the probes. Additionally, oxygen probes are highly sensitive to small changes in oxygen levels, unaffected by factors like blood volume, flow, local fluence, or wavelength-dependent scattering that can complicate Hb absorption-based methods, though temperature can influence lifetime of the probe.^101^ The direct and accurate pO2 measurement at a cellular or subcellular level is crucial for understanding the local oxygen environment, providing insights into cellular respiration, metabolism, and the physiological status of tissues.^102^ It is also important to note that pO2 and sO2 are not equivalent. sO2 varies with the pO2 in a nonlinear relationship that is affected by temperature, pH, and CO2. They can, however, be interconverted through the Hill equation.^98,103^
As a two-photon imaging technique, 2PLM of Oxyphor 2P and other probes offers submicron resolution, making it particularly suitable for pO2 measurements in capillaries, a feat challenging for other techniques. Its compatibility with two-photon fluorescence microscopy enables concurrent imaging of neuronal activity and microvasculature.^104,105^ As a result, 2PLM has been used extensively in investigating oxygenation change and its relationship with local neuronal activity in rodents under physiological and pathological conditions. For example, Li et al. found evidence for differences in oxygen extraction in cortical layers I-V of the whisker barrel cortex of awake mice with Oxyphor 2P and suggested that more homogeneous capillary flow and oxygenation correlates with increased oxygen extraction in deeper cortical layers, which is an adaptation to increased metabolism (Figure 5d).^106^ Şencan et al. observed that stimulus-induced changes in intravascular pO2 are conserved across cortical layers I–IV, suggesting a tightly-controlled neurovascular response to provide adequate O2 supply across cortical depth.^107^ Zhang et al. used 2PLM to measure locomotion effects on cerebral oxygenation in awake mice and found that breathing rate is a key modulator of cerebral oxygenation.^108^
Despite the unique strengths of 2PLM in imaging oxygenation in the brain, reliance on phosphorescent probes has intrinsic limitations. The main limitation is the slow measurement speed due to long phosphorescent lifetimes. Even with its improved speed, Oxyphor 2P requires 270 μs for an excitation-collection period which needs to be repeated 100–1000 times for averaging at a single point to collect enough photons and reach the SNR ratio needed for accurate pO2 quantification. As a result, the “imaging” is more like selected point measurements or at best coarse mapping at a rate of 0.1–1 sec per pixel, a few orders of magnitude slower than spectroscopy-based techniques. One recent study showed that the limit in speed could prevent observation of fast neurovascular coupling events.^109^ The need of injection or reinjection due to clearance is another limitation compared to label-free methods. Nevertheless, with high spatial resolution, comparable imaging depth to two-photon fluorescence of calcium indicators, and accurate measurements of pO2~, 2PLM is poised to make important contributions to the detailed understanding of neurovascular coupling. Recent demonstration with the use of two Pt porphyrin-based probes to simultaneously collect pO2 gradients from vessels and tissue in response to stimulus has achieved an imaging rate of 7 Hz.^102^ Further probe development and hardware optimization could potentially increase the imaging rate to >50 Hz, enabling direct correlation of local neuronal activity to oxygenation change at high speed.
Aside from the aforementioned optical imaging techniques, there are a few other potentially useful techniques for oxygen measurement in the brain. Transient Absorption Microscopy (TAM) is an emerging pump-probe microscopy technique that measures the nonlinear optical absorption of chromophores with two ultrashort laser pulses.^110^ Transient absorption signals can come from multiple processes, including two-photon absorption and excited state absorption (Figure 6a).^111^ By exploiting the transient absorption decay differences between HbO2 and HbR (Figure 6b), TAM offers a way to distinguish between these two forms of hemoglobin, providing a unique contrast for oxygen mapping. The initial demonstration in 2007 by Fu et al. showed that TAM can map blood vessels in live mouse ears and distinguish arterioles and venules in mouse ears down to 5–10 μm in diameter.^112,113^ However, the sequential imaging with two different pump/probe scenarios in these studies was inherently slow and thus unable to quantify sO2 due to motion. More recently, a ratiometric TAM method has achieved real-time sO2 quantification at the single RBC level in vitro (Figure 6c).^114^ TAM imaging of capillaries in the brain has been demonstrated at a maximum imaging depth of 250 μm (Figure 6d), but sO2 measurement was more challenging due to the need for additional calibration.^115^
The transient absorption of Hb is highly dependent on the wavelength pair used. To achieve sO2 quantification in the brain in vivo, the excited state decay differences between HbO2 and HbR need to be maximized through wavelength tuning. TAM’s ability to measure RBC velocity and hemoglobin concentration^116,117^, alongside its potential for fast sO2 imaging, positions it as a potential technique for comprehensive hemodynamic imaging in the brain. Its speed advantage over 2PLM and compatibility with two-photon fluorescence microscopy could facilitate detailed studies of oxygen delivery in relation to neuronal activity, albeit within the limitations of its imaging depth of a few hundred microns.
Third Harmonic Generation (THG) microscopy, a technique that generates signals at triple the frequency of the incident light, excels in imaging interfaces and regions with high optical inhomogeneity, such as blood vessels and myelin-rich axons in the brain (Figure 6e).^118,119^ This method can provide chemical specificity through resonant enhancement when an electronic transition of the molecule matches the harmonic frequency of the incident light. Ortas et al. hypothesized that if this resonant enhancement was exploited near the Soret band of Hb (415–430 nm), it would be sensitive to sO2.^118^ They used two lasers at 1300 and 1045 nm to generate THG and third-order sum frequency generation (TSFG) signals at 373, 401, and 433 nm. TSFG is similar to THG in that it is the signal from the sum of three photon energies, but the photons are of different wavelengths. By detecting multiple channels at once, the authors were able to detect oxygenation-dependent Hb signal in moving RBCs in the zebrafish brain (Figure 6f, g). However, sO2 quantification was not attempted. THG microscopy is a useful brain imaging technique due to its superior imaging depth over two-photon fluorescence.^120^ It remains to be seen whether resonance-enhanced THG could be used for mouse brain imaging, where scattering will be stronger than in the zebrafish brain. sO2 quantification could be challenging due to strong reabsorption of generated THG near the Soret band.
Raman spectroscopy of Hb has been extensively studied since the early 1970’s when several groups reported that some of the Hb vibrational peaks shifted frequency with oxygenation change.^121^ These studies utilized resonance Raman spectroscopy (RRS), where the incident laser energy coincides with a strong absorption band of Hb, which increases the Raman cross section and selectively enhances certain spectral features.^122^ HbO2 has several vibrational peaks at 1223, 1375, 1585, and 1638 cm^−1^ that are shifted to 1201, 1355, 1552, and 1602 cm^−1^, when deoxygenated. By comparing the intensity of the higher wavenumber HbO2 peak to the sum intensity of the HbO2 and HbR peak pair, a qualitative estimation of sO2 can be determined (Figure 6i).^123^ RRS has been used for a few in vivo oxygen saturation studies in a variety of tissues with good agreement with traditional oximetry methods.^124,125^ It has also been used non-invasively in animal models and even in neonates.^126,127^ Spatially resolved Raman microscopy for sO2 determination has been achieved in the brain in vivo by Brazhe et al.^123^ They used RRS to visualize the oxygenation in arterioles and venules in anesthetized mouse cortices with a 532 nm excitation to minimize photodamage (Figure 6j). However, the lack of spatial and temporal resolution is a major concern for in vivo measurements with RRS. Unfortunately, typical Raman spectral acquisition takes seconds, even slower than 2PLM.
Another implementation of Raman spectroscopy is coherent anti-Stokes Raman scattering (CARS), which is a nonlinear process and therefore possesses good 3D imaging capabilities and potentially deeper imaging depth at a few hundred micrometers.^128^ Using laser wavelengths of 710 and 950 nm, Rinia et al. demonstrated the feasibility of CARS for in vitro sO2 measurements of Hb solution with good accuracy (Figure 6i). However, interpretation of CARS spectra can be complicated due to the non-resonant background, which makes in vivo quantitative measurements difficult. An in vivo sO2 measurement has yet to be demonstrated.
Oxygen is crucial for brain metabolism and function, with its delivery and consumption tightly coupled to neural activity. The ability to image oxygen levels in the brain is vital for investigating neurovascular coupling and for comprehending and managing a wide array of neurological conditions and diseases. Optical imaging techniques, benefiting from Hb’s distinct optical absorption, play essential roles in measuring sO2 in the brain. These methods leverage the modified Beer-Lambert law at two or more wavelengths to determine HbO2 and HbR concentrations or change in concentrations. However, absolute concentration measurement is challenging due to unknown scattering contributions to the detected signals. Fortunately, sO2 calculation only requires the ratio of HbO2 and HbR.
Different wavelength regions have been used, depending on the employed technique. The relatively lower absorption and scattering enables fNIRS imaging of the brain at 2–3 centimeters deep, suitable for measurement of cortical activity in the human brain. However, because diffuse light is measured, fNIRS has a poor spatial resolution of several millimeters or worse. Thus, it is not suitable for microvascular imaging. The main advantage of fNIRS is its portability, allowing the study of brain function for human subjects freely moving and interacting with people and environments, and recording brain activities in children and infants. In another part of the light spectrum, visible wavelengths give strong absorption contrasts and sharp absorptive features but have shallower penetration depth. Both PAI and OCT use visible wavelengths to distinguish HbO2 and HbR. OCT relies on absorption measurement from ballistically scattered light and thus has limited imaging depths in the brain (<400 μm), but its micron-level spatial resolution enables single capillary sO2 imaging. The main applications of OCT to date have focused on retinal oximetry in both rodents and humans, where scattering is not a major concern, whereas brain studies have been sparse. PAI uses acoustic detection and thus overcomes the limitation of ballistic light scattering. Acoustic waves have longer wavelengths and significantly reduced scattering compared to light waves. More importantly, the detection wavelength is easily scalable, depending on the transducer center frequency. This unique feature facilitates multiscale PAM imaging with widely tunable resolution and imaging depth. AR-PAM achieves a lateral resolution from tens to hundreds of micrometers, with a depth of penetration from millimeters to a couple centimeters. To observe capillaries, OR-PAM uses optical focusing to achieve much finer resolution at the expense of limited imaging depth of 1 mm. The versatility of PAI enables brain studies in both rodents and larger animals, including humans. It is most suitable for fast imaging of hemodynamics in larger vessels in various animal models. However, to understand neurovascular coupling, it is important to monitor neuronal activity at the same time. Combining PAI with other optical techniques for neuronal imaging could be challenging. However, with the development of genetically encoded photoacoustic probes^122,123^, it might be possible to have simultaneous hemodynamic and neuronal imaging in the future. It remains to be seen if this approach can be applied to OR-PAM to study detailed neurovascular coupling in the cortex.
On a finer spatial scale, 2PLM of phosphorescence dyes such as Oxyphor 2P is very useful. It has intrinsically high spatial resolution (<1 μm), important for resolving capillary pO2. Additionally, the ability to probe tissue pO2 directly with high accuracy is a unique feature among all optical imaging techniques. The imaging depth, however, is limited to 600 μm. 2PLM will continue to play a critical role in unveiling local oxygen metabolism and the detailed mechanism of neurovascular coupling. However, its slow imaging speed is a major roadblock to many applications that require dense sampling of capillaries and fast imaging in response to neuronal activity. Other point scanning imaging modalities including TAM, Raman, and THG could potentially offer similar or better oxygen imaging capability. In particular, as an absorption-based technique, TAM could potentially provide orders of magnitude faster sO2~ imaging at similar imaging resolution as 2PLM. However, many refinements are needed for these new techniques to be applied to the brain in vivo. Given the importance of understanding oxygen metabolism and neurovascular coupling in the brain, it is expected that many more technical developments and further refinement will continue to flourish in the future. No single technique will be able to address all the measurement challenges, thus a fusion of imaging techniques and tailoring of techniques to specific problems will be essential to future brain imaging applications.