Intrinsic, widefield optical imaging of hemodynamics in rodent models of Alzheimer’s disease and neurological injury

Christian Crouzet; Thinh Phan; Robert H. Wilson; Teo Jeon Shin; Bernard Choi

doi:10.1117/1.NPh.10.2.020601

2 May 2023 Intrinsic, widefield optical imaging of hemodynamics in rodent models of Alzheimer’s disease and neurological injury

Christian Crouzet, Thinh Phan, Robert H. Wilson, Teo Jeon Shin, Bernard Choi

Author Affiliations +

Neurophotonics, Vol. 10, Issue 2, 020601 (May 2023). https://doi.org/10.1117/1.NPh.10.2.020601

Abstract

The complex cerebrovascular network is critical to controlling local cerebral blood flow (CBF) and maintaining brain homeostasis. Alzheimer’s disease (AD) and neurological injury can result in impaired CBF regulation, blood–brain barrier breakdown, neurovascular dysregulation, and ultimately impaired brain homeostasis. Measuring cortical hemodynamic changes in rodents can help elucidate the complex physiological dynamics that occur in AD and neurological injury. Widefield optical imaging approaches can measure hemodynamic information, such as CBF and oxygenation. These measurements can be performed over fields of view that range from millimeters to centimeters and probe up to the first few millimeters of rodent brain tissue. We discuss the principles and applications of three widefield optical imaging approaches that can measure cerebral hemodynamics: (1) optical intrinsic signal imaging, (2) laser speckle imaging, and (3) spatial frequency domain imaging. Future work in advancing widefield optical imaging approaches and employing multimodal instrumentation can enrich hemodynamic information content and help elucidate cerebrovascular mechanisms that lead to the development of therapeutic agents for AD and neurological injury.

1. Introduction

Although the brain is only 2% of the total body weight, it utilizes 20% of the heart’s cardiac output and 20% of the body’s oxygen and glucose supply under healthy conditions.¹ To ensure the brain receives adequate blood flow and energy, the cerebrovascular network modulates its vessel diameter to control local cerebral blood flow (CBF) and maintain brain homeostasis.²^,³ The cornerstone of these dynamics is the neurovascular unit (NVU). The NVU comprises different cell types, including endothelial cells, mural cells, glial cells, and neurons, which work together to regulate CBF and maintain the blood–brain barrier (BBB).⁴ However, neurodegeneration, which occurs in Alzheimer’s disease (AD) and neurological injury, can result in impaired CBF regulation, BBB breakdown, neurovascular dysregulation, and ultimately, impaired brain homeostasis.²^,⁵ To investigate these dynamics in rodent models, studies have used microscopic, mesoscopic, and macroscopic imaging approaches.

Microscopic approaches, such as two-photon excitation fluorescence imaging, provide exquisite cellular resolution that can examine the interplay between individual cell-types of the NVU. These approaches typically perform measurements localized to the first $1 {mm}^{3}$ of cortical brain tissue. They can provide anatomical and physiological views of the vasculature, CBF, and oxygen tension, providing mechanistic information at the cellular level. However, these approaches are confined to a small field of view (FOV), making examining connectivity between different brain regions challenging. On the other end of the spectrum, macroscopic approaches, such as positron emission tomography and magnetic resonance imaging (MRI), can probe the entire brain and are commonly used clinically in humans. These approaches enable the segmentation and comparison between the dynamics of CBF, cerebral metabolic rate of oxygen ( ${CMRO}_{2}$ ), and cerebral metabolic rate of glucose in different brain regions.⁶^–⁹ However, macroscopic approaches are expensive and lack contrast mechanisms that simultaneously detect a wide range of functional parameters.¹⁰

Mesoscopic approaches interrogate spatial scales that fall between the microscopic and macroscopic levels. For example, widefield optical imaging is a mesoscopic approach with an FOV ranging from millimeters to centimeters and probes up to the first few millimeters of rodent brain tissue. Widefield optical imaging approaches have measured both hemodynamic and neural activity. In this review, we will focus on hemodynamic activity but great reviews discuss widefield optical imaging in the context of examining neural activity.¹¹^–¹³ Widefield optical imaging approaches use different sources of contrast that enable the measurement of multiple hemodynamic parameters, such as CBF, oxygenation, and metabolism. Furthermore, measuring the entire cortical surface enables multiple FOVs to be compared, including different brain regions and vascular compartments. These advantages are critical to the advancement and implementation of widefield optical imaging technologies to monitor cerebral hemodynamic information.

This review first gives an anatomical and physiological overview of the cerebrovasculature in rodents. Next, we describe the principles of three widefield optical imaging approaches that can measure cerebral hemodynamics: (1) optical intrinsic signal imaging (OISI), (2) laser speckle imaging (LSI), and (3) spatial frequency domain imaging (SFDI). Then, we discuss measurements that can be performed with these approaches, such as resting-state dynamics, neurovascular coupling (NVC), and cerebrovascular reactivity (CVR). Next, we discuss how each technology has been used to measure physiological changes in AD and neurological injuries. Finally, we discuss the future directions for widefield optical imaging technologies in measuring cerebral hemodynamics.

2. Cerebrovascular System

The cerebrovasculature delivers nutrients, such as oxygen and glucose, to the brain. These nutrients enter the surrounding brain tissue through the BBB, primarily in capillaries. Similarly, metabolic waste products, such as carbon dioxide ( ${CO}_{2}$ ), cross the BBB into the bloodstream to be cleared from the body through the lungs, kidneys, and liver. Oxygen and ${CO}_{2}$ rapidly diffuse across the BBB. Larger nutrients and waste products cross the BBB using regulated transport receptors.¹⁴ The importance of these basic cerebral functions is exemplified by the fact that the brain receives 20% of the cardiac output and consumes 20% of the body’s supply of oxygen and glucose.¹⁵ This high metabolic activity is facilitated by the large cerebral microvascular blood volume fraction ( $\sim 16 %$ in mice).¹⁶

Blood flow to the mouse brain originates from the internal carotid and vertebral arteries.¹⁷ In contrast to the circle of Willis in humans, in mice, the cerebral arteries of Willis do not form a complete loop.¹⁸ Instead, two distinct domains are supplied by either the internal carotid or the basilar artery. Due to the significant role that these large arteries play in modulating cerebrovascular resistance,¹⁹ they have a prominent role in the overall regulation of CBF. From the circle of Willis, three main arteries (anterior, middle, and posterior cerebral arteries) arise that progressively branch into smaller pial arteries that are present on the surface of the brain. The pial vessels then plunge into the brain as penetrating arterioles that subsequently branch into precapillary arterioles and capillaries, which are critical in nutrient delivery and waste removal.

To efficiently deliver nutrients to and remove waste products from the brain, a process known as neurovascular coupling (NVC) occurs. In response to increased local neural activity, CBF rapidly increases,²⁰^,²¹ which brings nutrients (i.e., oxygen and glucose) to meet the increased metabolic demand. The supply is large enough to ensure that neurons distal to individual cerebral blood vessels receive the necessary metabolites via diffusion along a concentration gradient.²² Local control of CBF is modulated by the cells that comprise the NVU: endothelial cells, pericytes, vascular smooth muscle cells (VSMCs), astrocytes, microglia, and neurons.⁵^,²⁰^,²³^,²⁴ More specifically, as local neuronal activity increases, surrounding arterioles vasodilate. An increase in local CBF occurs as a retrograde signaling cascade propagates to upstream arteries, including the pial arteries.²⁵^–²⁷ Gap junctions facilitate rapid retrograde propagation, which is proposed to occur via K+ ions released during neural activity,²⁸ of the signal between endothelial cells,²⁹ and surrounding pericytes and VSMCs subsequently relax, leading to local vasodilatation. Vasoactive mediators, such as nitric oxide, epoxyeicosatrienoic acids (EETs), and prostaglandin E2 (PGE2), act in concert to achieve vasodilation during NVC.²⁸^,³⁰ Interneurons may also modulate vessel diameter.³¹^,³²

In addition to NVC, cerebrovascular health and function can be studied via CVR tests to external stimuli. For example, pharmacological procedures can be performed to induce local vasodilatation, including superfusion of acetylcholine, which causes the release of endothelial nitric oxide³³ and administration of nitric oxide donors (i.e., S-nitroso-N-acetylpenicillamine) to the exposed mouse cortex.³⁴ Another commonly used approach to study CVR is hypercapnia, which uses increased inspired ${CO}_{2}$ to stimulate vasodilation.³⁵ In contrast, the superfusion of amyloid-beta ( $A β$ ) leads to impaired CVR.³⁵ To quantify CVR, CBF responses to these external stimuli are typically measured using optical technologies, such as laser Doppler flowmetry³⁴^,³⁵ and LSI,³⁶^,³⁷ and medical imaging technologies such as blood-oxygen-level-dependent functional MRI (fMRI)³⁸ is used.

The general NVU composition varies at different levels of the cerebral vasculature.²^,²⁸ Cerebral arteries contain multiple layers of VSMCs to control CBF. Pial arterioles on the cortical surface have a thinner layer of VSMCs. In contrast, penetrating arterioles have a single VSMC layer that adjoins the endothelial cells. A perivascular space separates the basement membrane of these arterioles from the endfeet of astrocytes. Macrophages and leptomeningeal cells surround these arterioles. With smaller intraparenchymal arterioles, the perivascular space is no longer present, and astrocytes directly connect with the basement membrane. As penetrating arterioles branch into smaller intraparenchymal arterioles, transitional pericytes may be present with properties similar to both pericytes and VSMCs. Pericytes replace VSMCs at the capillary level, and neurons are close to a vascular basement membrane that consists of both pericytes and endothelial cells. Similar to arterioles, venules have a perivascular space that separates the vascular basement membrane from astrocytes. Larger veins have a thinner and flatter layer of VSMCs than arteries and a progressively larger perivascular space separating the vascular basement membrane from astrocytes. Schaeffer and Iadecola²⁸ provide an excellent description of what they call the neurovascular complex, which consists of multiple versions of the NVU along the length of the cerebrovasculature. For example, they summarize RNA sequencing work that identified seven endothelial cell clusters in the NVU. These local variations in NVU enable efficient and robust communication throughout the cerebral vasculature and cerebrovascular dynamics throughout the entire brain.

The arteries and arterioles also play a significant role in regulating CBF in response to changes in arterial pressure. The pulsatility of blood due to the beating heart is dampened by the arteries and arterioles, resulting in minimal pulsatility at the capillary level. Also, cerebral autoregulation processes maintain a stable CBF over a wide range of arterial blood pressures. The myogenic response of VSMCs enables vasoconstriction at high arterial pressure and vasodilation at low arterial pressure, to achieve sufficient (but not excessive) CBF for different intravascular pressures.³⁹

3. Core Technologies

3.1.

Sources of Contrast

OISI, LSI, and SFDI use sources of optical contrast that are critical to measuring hemodynamics: (1) optical absorption, (2) optical scattering, and (3) speckle contrast. Since each source of contrast varies depending on the wavelength, care should be taken when selecting the wavelength(s) of light for device builds.

Absorption occurs when a chromophore (e.g., oxyhemoglobin) absorbs a photon of light. Scattering, which is high in biological tissue at visible and near-infrared wavelengths, occurs when a photon changes direction due to local refractive index mismatches. The absorption coefficient ( $μ_{a}$ ) is defined as the probability of absorption per unit distance, and the scattering coefficient ( $μ_{s}$ ) is defined as the probability of scattering per unit distance. The absorption due to hemoglobin is essential to measure cerebral hemodynamics. Visible to near-infrared (400 to 1000 nm) wavelengths have a highly variable absorption spectrum. The wavelengths from $\sim 400$ to 600 nm provide excellent signal-to-noise to measure hemoglobin changes, whereas wavelengths from $\sim 650$ to 1000 nm allow for increased depth penetration due to the near-infrared optical window. The near-infrared optical window exists due to an $\sim 100$ -fold decrease in hemoglobin absorption from visible to near-infrared light and is commonly used in diffuse optics.

The reduced scattering coefficient ( $μ_{s}^{'}$ ) is commonly used in place of $μ_{s}$ to describe tissue optical properties. It comprises $μ_{s}$ and anisotropy (g) of the tissue, where $μ_{s}^{'} = (1 - g) μ_{s}$ . $g$ is the average cosine of all scattering angles and describes how forward-scattering each scattering event is. A typical value for $g$ is $\sim 0.9$ in biological tissue. $μ_{s}^{'}$ is used in the diffuse regime (i.e., $μ_{a} ≪ μ_{s}^{'}$ ) and describes the propagation of photons experiencing multiple scattering events. Unlike absorption, which is highly variable in the visible to near-infrared regime, scattering decreases with increasing wavelength as a power law function.

Speckle contrast is the last source of optical contrast we discuss. It can be obtained from analyzing the speckle pattern resulting from coherent light interference. The speckle contrast describes how much the speckle pattern fluctuates over time. The faster the speckle pattern fluctuates, the lower the speckle contrast, whereas the slower the speckle pattern fluctuates, the higher the speckle contrast. This information is used to obtain quantitative blood flow information. The three sources of optical contrast (absorption, scattering, and speckle contrast) are critical and instrumental for OISI, LSI, and SFDI to provide cerebral hemodynamic information. Next, we discuss the basic principles associated with each imaging approach.

3.2.

Optical Intrinsic Signal Imaging

OISI or intrinsic signal optical imaging was developed by Grinvald et al.⁴⁰ to visualize and quantify information related to neural activity. OISI measures changes in the reflected light from tissue associated with absorption and scattering dynamics. As the name suggests, OISI does not require using exogenous probes typically required for extrinsic fluorescence (e.g., calcium or voltage-sensitive dyes) or phosphorescence (e.g., oxygen tracing). The basic principle of OISI is that a perturbation can induce a change in reflectance (measured with a camera or sensor array) that gives a map of neural activities.⁴⁰ To assess these dynamics, a typical hardware setup for OISI includes a camera sensor, a light source with one or more wavelengths, and a computer for triggering, acquisition, and processing [Fig. 1(a)].¹²^,⁴¹^–⁴³ Light sources for OISI can use light-emitting diodes (LEDs) or a broadband light source (such as a lamp) with a filtered wheel.⁴³

Fig. 1

OISI instrumentation, processing scheme, and data quantification. (a) A typical OISI setup where a computer syncs the exposure time of the camera and switching of the LEDs. DM, dichroic mirror; CL, condenser lens; P, polarizer; LED, light-emitting diode; Cam, camera. (b) OISI processing pipeline from raw data with representative images of whisker stimulation on awake mice (adapted from Ref. 12). (Top) Using the original OISI approach, relative reflectance changes in % of 530 ( $\sim HbT$ ) and 630 nm ( $\sim HbR$ ) light during whisker stimulation are visualized (adapted from Ref. 12). (Bottom) Using spectroscopic OISI, relative changes in $C_{HbR}$ , $C_{HbO 2}$ , $C_{HbT}$ , and calcium indicator GCaMP fluorescence during whisker stimulation are visualized (adapted from Ref. 12).

Since its conception, OISI has been widely used in neuroimaging due to its ability to provide high spatial and temporal resolution maps of brain activity. The depth sensitivity of OISI is typically localized to the cortical layers, often through a cranial window preparation. In rodents, the procedure often entails a craniectomy to expose the cortical region of interest (ROI), followed by a glass coverslip sealant. OISI has also been used with thinned⁴⁴^,⁴⁵ and intact skull preparations on mice.⁴¹^,⁴⁶^,⁴⁷ Nevertheless, all methods require the retraction of the scalp and removal of the fascia layer before imaging.

The original method for OISI uses reflectance changes from the backscattered light to observe the spatiotemporal dynamics of functional activity [Fig. 1(b), top].⁴⁰ A key improvement involves spectroscopic analysis [Fig. 1(b), bottom], which enables the measurement of changes in chromophores, such as oxyhemoglobin ( ${HbO}_{2}$ ) and deoxyhemoglobin (HbR).⁴⁸ Spectroscopic OISI employs the modified Beer–Lambert law [Eq. (1)]:⁴⁹^,⁵⁰

Eq. (1)

I = I_{0} * \exp (- μ_{a} * X + G),

where

I

is the detected light intensity,

I_{0}

is the light source intensity,

X

is the total path length of light in tissue,

μ_{a}

is the absorption coefficient, and

G

is a geometry factor.⁴⁹ It should be noted that

X = (DPF) x

, where DPF is the differential path length, which is a scaling factor between total path length and the source–detector separation,

x

.¹² In near-infrared spectroscopy, the DPF is often used instead of the total path length since the source–detector separation is a known constant.⁵¹ Next, spectroscopic OISI often describes results to be relative to the initial time,

t_{0}

. Thus, Eq. (1) can be rewritten as

Eq. (2)

Δ μ_{a} (t, λ) = μ_{a} (t, λ) - μ_{a} (t_{0}, λ) = - \frac{1}{X (λ)} \ln (\frac{I (t, λ)}{I (t_{0}, λ)}) .

The complete derivation for Eqs. (2)–(4) is described in Ref. 12. If a minimum of two wavelengths ( $λ_{1}$ and $λ_{2}$ ) are used, and $μ_{a}$ is defined as the sum of products between the chromophore concentrations of interest (i.e., ${HbO}_{2}$ and HbR) and their associated extinction coefficients ( $ε$ ), the following set of equations are obtained:

Eq. (3)

Δ μ_{a} (t, λ_{1}) = ε_{{HbO}_{2}} (λ_{1}) Δ C_{{HbO}_{2}} (t) + ε_{H b} (λ_{1}) Δ C_{HbR} (t) Δ μ_{a} (t, λ_{2}) = ε_{{HbO}_{2}} (λ_{2}) Δ C_{{HbO}_{2}} (t) + ε_{Hb} (λ_{2}) Δ C_{HbR} (t) .

For Eqs. (2) and (3), X is required to estimate the change in concentrations of ${HbO}_{2}$ ( $Δ C_{HbO 2}$ ) and HbR ( $Δ C_{HbR}$ ). In recent studies, Monte Carlo simulations of photon migration in homogeneous media have been utilized to obtain estimates of $X$ .⁴⁴^,⁵² To run the simulation, researchers often use brain optical properties ( $μ_{a}$ and $μ_{s}^{'}$ ) from the literature.⁵³ Using literature values of $X$ , one can then achieve a wavelength-dependent linear system of equations with chromophore concentrations as unknowns. Theoretically, the system of equations can be solved algebraically for $N$ chromophores using $N$ wavelengths. However, a more common practice involves using least square fitting for the matrix form of the equations. This method requires at least $N + 1$ wavelengths to improve fitting accuracy:

Eq. (4)

[\begin{matrix} Δ μ_{a} (t, λ_{1}) \\ ⋮ \\ Δ μ_{a} (t, λ_{n}) \end{matrix}] = [\begin{matrix} ε_{{HbO}_{2}} (λ_{1}) & ε_{HbR} (λ_{1}) \\ ⋮ & ⋮ \\ ε_{{HbO}_{2}} (λ_{n}) & ε_{HbR} (λ_{n}) \end{matrix}] [\begin{matrix} Δ C_{{HbO}_{2}} (t) \\ Δ C_{HbR} (t) \end{matrix}] .

Using the above principles, OISI can measure absorption dynamics due to changes in blood volume, CBF, and tissue oxygenation in response to neuronal activation. In addition, OISI can also detect a change in reflectance associated with a change in scattering properties, which may be due to vascular and extracellular compartment changes, neuronal cellular expansion, and movement of water and ions.⁵⁴^,⁵⁵ Although OISI performs well at measuring relative changes in hemoglobin concentrations, it does not measure absolute hemoglobin concentrations well due to the inability of OISI to separate effects due to absorption and scattering. We discuss how SFDI overcomes this limitation later in this review.

3.3.

Laser Speckle Imaging

LSI, also known as laser speckle contrast imaging (LSCI) and laser speckle contrast analysis (LASCA), is a simple widefield imaging approach to measure relative blood flow. Fercher and Briers demonstrated the first use of LSI to visualize blood flow in biological tissue.⁵⁶ Twenty years after this initial demonstration, Dunn et al.⁵⁷ applied LSI to the brain and visualized CBF dynamics during cortical spreading depression (CSD). Since the publication of this seminal work, LSI has been applied to several neurologically relevant applications, such as stroke,⁵⁸^,⁵⁹ cardiac arrest,⁶⁰^–⁶² AD,⁶³ and aging⁶⁴ to assess CBF dynamics. To obtain blood flow information, LSI hardware requires a laser, camera, sample to image, and computer for data acquisition and analysis [Fig. 2(a)]. When coherent laser light illuminates a sample, a random interference pattern, called a speckle pattern, is produced on the camera sensor. The speckle pattern contains bright and dark pixel intensities due to light that travels different path lengths in the sample. The bright and dark pixel intensities on the camera sensor are locations where constructive and destructive interference occur, respectively. When scattering particles (e.g., RBCs) move in a sample, the speckle pattern fluctuates in time, which appears as intensity variations on the camera detector. These intensity variations can be used to obtain information regarding the motion (e.g., blood flow) of the scattering particles.

Fig. 2

LSI instrumentation, processing scheme, and data quantification. (a) A typical LSI setup where a computer controls the camera. LAS, laser; AL, aspheric lens; P, polarizer; and Cam, camera. LSI processing pipeline from raw data to blood flow maps using (left) spatial and (right) temporal processing algorithms. (b) Quantitation of LSI data. Choosing an ROI depends on the brain being investigated. Three different types of measurements: (top) NVC in response to hindpaw stimulation with quantitative parameters, slope, and peak response; (middle) CVR in response to ${CO}_{2}$ inhalation with quantitative parameters, $t_{50 rise}$ , AUC of relative CBF during hypercapnia, maximum change in relative CBF (rCBF), and $t_{50 fall}$ ; (bottom) CBF pulsatility at rest with peak and trough outlined, showing an amplitude can be measured.

An outline of traditional LSI data processing is shown in Fig. 2(a). A raw speckle image comprises a map of intensity fluctuations integrated over the camera’s exposure time (typically 1 to 10 ms for LSI). In areas of increased blood flow, the intensity fluctuations of the speckle pattern change more quickly, leading to a blurring of the raw speckle image. To quantify blood flow from the speckle pattern fluctuations, researchers have primarily used two processing algorithms to account for the spatial and temporal statistics of the speckle pattern. The spatial algorithm typically uses a $5 \times 5$ or $7 \times 7$ sliding window. The spatial standard deviation ( $σ_{s}$ ) and spatial mean ( ${⟨ I ⟩}_{s}$ ) are computed using the pixel intensities within the sliding window and assign the center pixel the spatial speckle contrast ( $K_{s}$ ). The sliding window slides throughout the entire image until the computation is complete. The temporal algorithm computes the temporal speckle contrast ( $K_{t}$ ) from a series of images. The temporal standard deviation ( $σ_{t}$ ) and temporal mean ( ${⟨ I ⟩}_{t}$ ) are calculated using the pixel intensities at each pixel location across a series of images. From the calculated speckle contrast, spatial or temporal, a relative blood flow can be calculated by taking the inverse of the correlation time ( $τ_{c}$ ) in Eq. (5) (i.e., blood flow $\propto 1 / τ_{c}$ ). $T$ is the exposure time of the camera and $β$ is a normalization factor to account for speckle averaging due to a mismatch of speckle size and detector size, polarization, and coherence effects. $β$ is normally assumed to be one but can be measured experimentally to obtain a $β$ -corrected speckle contrast, which we discuss later in this section:

Eq. (5)

K = \frac{σ}{⟨ I ⟩} = {[β (\frac{τ_{c}}{T} + \frac{τ_{c}^{2}}{2 T^{2}} (\exp (- \frac{2 T}{τ_{c}}) - 1))]}^{1 / 2} .

Our group and others have derived a simplified speckle imaging equation for calculating relative blood flow by imposing assumptions on Eqs. (5) and (6).⁶⁵^,⁶⁶ The speckle flow index (SFI) is proportional to blood flow and approximates $τ_{c}$ as $2 {TK}^{2}$ :

Eq. (6)

SFI = \frac{β}{2 T K^{2}} .

There are advantages and disadvantages when using the spatial and temporal algorithms to obtain speckle contrast. For example, the spatial algorithm has a higher temporal resolution but lower spatial resolution than the temporal algorithm. Importantly, due to the increased temporal resolution with the spatial algorithm, it can resolve pulsatile fluctuations in blood flow with each contraction of the heart.⁶²^,⁶⁷^–⁷⁰ Due to the principle of ergodicity, the spatial and temporal algorithms can approach the same quantitative result when a craniotomy is performed since the static scattering from the skull is removed. However, biological tissue still has static and dynamic scattering components, which makes it difficult to achieve ergodicity in biological tissue.⁷¹ Despite this, areas measured directly over a vessel may be ergodic if the scattering events only interact with that vessel. When the skull remains intact, the spatial algorithm produces an offset to the spatial speckle contrast due to the static scattering from the skull. The temporal algorithm is less susceptible to artifacts caused by the skull and minimally affects the temporal speckle contrast. Although the spatial and temporal algorithms can be used individually, spatiotemporal approaches that combine aspects of the spatial and temporal algorithms are commonly used.⁷² We discuss some of the limitations associated with LSI in the resting-state blood flow section and potential ways to overcome these limitations. Despite some of the limitations associated with traditional LSI, it has been extensively used in diseased and healthy rodent brains. In this review, we highlight areas that LSI has been applied.

3.4.

Spatial Frequency Domain Imaging

SFDI⁷³ is a relatively new technology designed to quantitatively map tissue absorption and scattering over a wide FOV.⁷⁴ To perform this measurement, a typical hardware setup for SFDI includes a camera sensor, a light source with one or more wavelengths, a digital micromirror device (DMD), and a computer for triggering, acquisition, and processing [Fig. 3(a)].⁷⁵^–⁷⁷ SFDI uses a DMD to project spatially modulated patterns (either from a broadband source or multiple discrete wavelengths) onto the tissue and detects the backscattered light with a camera. The projected light patterns are typically sinusoidal, with different spatial frequencies. However, square-wave patterns have also been used in place of sinusoids to enable more rapid data acquisition.⁷⁸ Each pattern projected onto the tissue undergoes amplitude attenuation and blurring due to tissue absorption and scattering, respectively. Since each spatial frequency has a unique combination of attenuation and blurring during its interaction with the tissue, using multiple different spatial frequencies enables the separation of the tissue $μ_{a}$ from $μ_{s}^{'}$ .

Fig. 3

SFDI instrumentation and processing scheme. (a) A typical SFDI setup where a computer controls the camera, LEDs, and DMD. LED, light-emitting diode; CL, condenser lens; DM, dichroic mirror; DMD, digital micromirror device; P, polarizer; L, lens; and Cam, camera. (b) SFDI processing pipeline from raw data to tissue optical property maps and scatter-corrected chromophore concentrations. Raw intensity images at each wavelength are collected at two or more spatial frequencies, $f_{0}$ and $f_{x}$ . Images with each $f_{x}$ are taken at three different phases (0 deg, 120 deg, and 240 deg) of the spatially modulated pattern. Next, the diffuse reflectance is calculated at each spatial frequency and wavelength. Using light-transport models, the optical properties ( $μ_{a}$ and $μ_{s}^{'}$ ) as a function of wavelength $λ$ are obtained. The scatter-corrected absorption can then be used to obtain chromophore concentrations (i.e., $C_{HbO 2}$ , $C_{HbR}$ , $C_{HbT}$ ) and tissue oxygenation ( ${StO}_{2}$ ).

Figure 3(b) shows the SFDI processing pipeline from raw data to tissue optical property maps and scatter-corrected chromophore concentrations. To process the raw SFDI data, the images collected at each spatial frequency are first demodulated to obtain the demodulated intensity image $I_{AC, tissue}$ :

Eq. (7)

I_{AC, tissue} (f_{x}) = \frac{\sqrt{2}}{3} [{(I_{1} - I_{2})}^{2} + {(I_{2} - I_{3})}^{2} + {(I_{3} - I_{1})}^{2}],

where

I_{1}

,

I_{2}

, and

I_{3}

are the two-dimensional raw backscattered intensity maps collected with three different spatial phase shifts of the SFDI pattern (typically 0, 120 deg, and 240 deg). The outputs from the demodulated images are the measured AC component of the reflectance (

I_{AC, tissue} (f_{x})

) maps at each spatial frequency. SFDI measurements also usually include an unstructured (

f_{x} = 0

) backscattered intensity image (which does not require demodulation since it does not contain a spatial pattern). Next, the demodulated images are calibrated against a corresponding measurement

I_{AC, phantom} (f_{x})

of a tissue-simulating phantom with known optical properties, using Eq. (8):

Eq. (8)

R_{d_{tissue}} (f_{x}) = [R_{d_{phantom}} (f_{x})] [\frac{I_{AC, tissue} (f_{x})}{I_{AC, phantom} (f_{x})}] .

The resulting quantity $R_{d_{tissue}} (f_{x})$ is the diffuse reflectance of the tissue for each spatial frequency $f_{x}$ of the incident light. $R_{d_{tissue}} (f_{x})$ is an intrinsic property of the tissue, independent of the components of the measurement setup, that varies from 0 (for a completely absorbing medium) to 1 (for a medium with no absorption or scattering-related losses). Two-dimensional maps of $R_{d_{tissue}} (f_{x})$ over the camera’s FOV are typically obtained at multiple wavelengths. Note that obtaining a quantitatively accurate value of $R_{d_{tissue}} (f_{x})$ depends on knowing $R_{d_{phantom}} (f_{x})$ , which is typically obtained using a separate paradigm, such as using an integrating sphere.⁷⁹

At each wavelength, the $R_{d_{tissue}} (f_{x})$ curve is fit with a mathematical model of light transport in tissue as a function of spatial frequency and tissue optical properties. This model is typically based on Monte Carlo simulations or the diffusion approximation to the radiative transport equation. The best fit of the model to $R_{d_{tissue}} (f_{x})$ at each wavelength is used to extract the corresponding $μ_{a}$ and $μ_{s}^{'}$ values at that wavelength. These spectrally resolved $μ_{a}$ and $μ_{s}^{'}$ values are fit to physiological models of tissue absorption and scattering to extract concentrations of different tissue components. For instance, in the case of absorption in brain tissue, the $μ_{a}$ spectrum can be approximated as a linear combination of the molar extinction coefficients of oxygenated and deoxygenated hemoglobin, shown in Eq. (9):

Eq. (9)

μ_{a} (λ) = 2.303 [ε_{HbO 2} (λ) C_{HbO 2} + ε_{HbR} (λ) C_{HbR}] .

As stated in the OISI section, $ε_{{HbO}_{2}}$ and $ε_{HbR}$ are the molar extinction coefficients of oxygenated and deoxygenated hemoglobin, respectively, weighted by the concentrations of ${HbO}_{2}$ and HbR. Thus, the best fit of this model to the $μ_{a} (λ)$ spectrum yields the respective concentrations of ${HbO}_{2}$ ( $C_{HbO 2}$ ) and HbR ( $C_{HbR}$ ) of oxygenated and deoxygenated hemoglobin in the tissue. The total hemoglobin concentration ( $C_{HbT}$ ) of the tissue is thus given by the summation $C_{HbT} = C_{HbO 2} + C_{HbR}$ , and the tissue oxygenation ( ${StO}_{2}$ ) is given by the equation ${StO}_{2} = 100 (C_{HbO 2} / C_{HbT})$ .

In addition, since illumination at different spatial frequencies is the frequency-domain analog of diffuse reflectance measurements at multiple spatial source–detector separations,⁸⁰ SFDI effectively interrogates a volume of tissue whose depth depends on the spatial frequency. This feature of SFDI enables the separation of the contributions of different tissue layers to the detected signal.⁸¹ It can facilitate three-dimensional tomographic reconstruction of subsurface perturbations to tissue optical properties.⁸¹^,⁸² Studies have also shown that using a combination of different orientations of structured light patterns⁸³ or employing a wider range of spatial frequencies⁸⁴ can provide sensitivity to higher-order parameters related to the directionality of the scattering of light by the tissue.

4. Types of Measurements

We discuss three measurements that widefield optical imaging approaches can perform: (1) resting-state, (2) NVC, and (3) CVR. These three measurements are used to assess cerebrovascular health and function. We discuss these measurements in the context of awake and anesthetized rodents.

4.1.

Resting-State

Baseline or resting-state measurements can be used to examine unperturbed changes in hemodynamics in the awake or anesthetized state. Here, we highlight blood flow and functional connectivity as areas where widefield optical imaging approaches have significantly progressed and how quantitative information can be obtained from these areas.

4.1.1.

Blood flow

Traditional LSI is only able to measure relative blood flow changes accurately. Resting-state blood flow has been challenging to measure due to factors that influence the speckle contrast, such as day-to-day instrument variability,⁸⁵ static scattering of the skull,⁸⁵^,⁸⁶ and tissue optical properties.⁸⁷^,⁸⁸ Equation (5) in the LSI section shows the direct relationship between the speckle contrast ( $K$ ) and the decorrelation time ( $τ_{c}$ ,) where it is accepted that $CBF \sim 1 / τ_{c}$ .⁸⁵^,⁸⁶^,⁸⁹ However, $β$ in Eqs. (5) and (6) is a term that depends on the differences in day-to-day instrument variability, experimental environment, differences in detector-to-speckle size ratio, and polarization. However, the $β$ term is normally assumed to be one, which does not take into account day-to-day instrument variability. To account for day-to-day instrument variability, $β$ can be measured on each imaging day. Although $β$ is fairly easy to measure, it is rarely performed. $β$ can be measured by imaging an object where $τ_{c} ≫ T$ , which is generally true for a static phantom. The measured $β$ can be inserted into Eqs. (5) or (6) to obtain a more accurate blood flow measurement than assuming a $β$ of one. This approach theoretically enables a comparison between animals and imaging days if the optical properties of the samples are similar.

Parthasarathy et al.⁸⁵ implemented multiexposure speckle imaging (MESI) by deriving a new speckle equation and imaging system in the presence of static scattering (e.g., skull) that accounts for $β$ , the fraction of dynamically scattered light, ergodicity of the sample, and noise of the imaging system. This methodology enables comparison across several days of imaging. Furthermore, MESI is theoretically more sensitive to blood flow changes than traditional LSI due to the multiple exposure times used. Although the technique works well, it has not gained widespread use, most likely due to the complexity of implementing the MESI system correctly compared with traditional LSI. Recently, an all-fiber-based illumination system was developed to reduce the size and complexity of traditional MESI systems.⁹⁰ This design showed good agreement with standard MESI and may increase the adoption of MESI.⁹⁰ In addition to MESI, advances have been made to better understand and improve the mathematical models used to describe laser speckle fluctuations in dynamic light scattering, including LSI. Postnov et al.⁹¹ performed a thorough characterization to investigate the type of motion (i.e., unordered or ordered) and scattering regime (i.e., single or multiple) used in dynamic light scattering. Importantly, this study revealed that the type of model used to obtain blood flow can vary widely over an FOV, especially for different vessel sizes, and can drastically impact the resultant blood flow measurement.⁹¹ Specifically, LSI can underestimate the decrease in CBF associated with ischemic stroke.⁹¹ A subsequent study⁹² further expanded on this work by deriving new models to be used directly for LSI that account for different combinations of scattering and motion type. Using the incorrect model can lead to errors of 300%.⁹² In addition to issues with modeling dynamic light scattering, optical properties also impact speckle contrast.⁸⁸ To account for tissue optical properties in the speckle contrast, we have combined LSI with SFDI to measure tissue optical properties.⁷⁷^,⁸⁷^,⁹³ The measured optical properties enable an optical property-corrected blood flow.⁷⁷^,⁹³ Implementing these methods can lead to greater utility of LSI in longitudinal brain imaging studies.

When measuring resting-state CBF, it is important to consider the effects of anesthesia. Most inhaled anesthetics increase CBF. One example is isoflurane, an inhaled anesthetic commonly used during rodent surgeries and optical imaging. CBF dramatically increases via vasodilation under isoflurane anesthesia⁹⁴ compared with the awake state, in addition to impairing cerebral autoregulation,⁹⁵^–⁹⁷ and can lead to a diffuse NVC response. Alpha-chloralose is another common anesthetic used to assess NVC and cerebral autoregulation since both are preserved. However, resting-state CBF is drastically reduced in rodents that receive alpha-chloralose as an anesthetic.⁹⁸^–¹⁰⁰ In addition, alpha-chloralose is a terminal anesthetic, which is not approved for survival experiments.¹⁰¹ Finally, ketamine-xylazine is a common anesthetic given i.p. in rodents to assess NVC and cerebral autoregulation. In comparison to isoflurane and alpha-chloralose, ketamine-xylazine minimally alters resting-state CBF.⁹⁸ In all, anesthetics exert variable effects on resting-state CBF that differ from awake, resting-state CBF. If possible, resting-state CBF measurements with LSI should be made under awake conditions.

4.1.2.

Functional connectivity

Functional connectivity is the temporal correlation of neuronal activity between anatomically separated brain regions.¹⁰² The concept stems from image analysis using fMRI scans during rest, thus named resting-state functional connectivity. The presence of correlation during the resting state suggests that studies of its activities can be achieved without perturbations, such as stimulations or cognitive tasks.⁴¹ Widefield optical imaging approaches have enabled the analysis of functional connectivity in mice.⁴¹^,¹⁰³ OISI was the first widefield imaging technique used to obtain resting-state correlation maps over the mouse cortex.⁴¹ The method typically entails choosing a small ROI within a functional region of the brain (e.g., visual or motor) on the resulting hemoglobin concentration maps (e.g., ${HbO}_{2}$ and HbT) as a seed pixel to obtain a time-traced signal.⁴¹^,⁴⁶^,¹⁰³ In addition, individual OISI wavelengths have been selected to perform functional connectivity analyses and are quantitatively comparable to ${HbO}_{2}$ and HbT analyses.¹⁰⁴ LSI was also used to obtain time-traced signals based on blood flow maps but has not seen much use in functional connectivity due to its inherent noisy nature and requires both spatial and temporal filtering.¹⁰³ It is common practice to filter the time traces for the specific frequency bands of interest during the pixelwise time trace extraction, including infraslow (0.009 to 0.08 Hz), intermediate (0.08 to 0.4 Hz), and high (0.4 to 4.0 Hz) frequency bands.⁴⁶ From these filtered time traces, a temporal correlation analysis is performed in a pairwise manner between a seed pixel and all other pixels over the sampled brain region. The correlation analysis yields a map of correlation coefficients (or fisher Z scores) relative to the seed pixel, which is the functional connectivity map.

When performing functional connectivity analysis, one important feature is the type of anesthesia used. For example, a study that used a transgenic mouse model with the calcium indicator, GCaMP6, showed that the spatial extent of seed-based functional connectivity vastly differs depending on the state of consciousness (i.e., awake versus anesthetized).⁴⁶ Unfortunately, previous functional connectivity studies have used various anesthetic agents, which makes it difficult to compare findings. White et al. selected ketamine-xylazine as its anesthetic for OISI and showed it is appropriate to analyze signals in the infraslow band (0.009 to 0.08 Hz).⁴¹ In a study of functional connectivity with fMRI in rats, isoflurane reduced cross-correlation coefficients in comparison to alpha-chloralose and medetomidine.¹⁰⁵ However, a subsequent fMRI study in mice showed that interhemispheric functional connectivity was diminished for alpha-chloralose and urethane in comparison to isoflurane and awake groups. Despite these studies, a recent study compared the effects of five different anesthetics: avertin, ketamine-xylazine, urethane, chloral hydrate, and isoflurane.¹⁰⁴ This study showed that avertin and ketamine-xylazine produce robust and repeatable resting-state functional connectivity results, whereas isoflurane and chloral hydrate did not. Based on widefield functional optical imaging studies that have measured functional connectivity, if an anesthetic is needed, avertin or ketamine-xylazine is recommended. However, performing functional connectivity analysis in the awake state is optimal.

4.2.

Neurovascular Coupling

In the normal brain, NVC refers to the local increase in CBF in response to neuronal activity.¹⁰⁶^–¹⁰⁸ NVC is a critical control mechanism to supply sufficient nutrients to neurons during activation.²¹ In the context of widefield functional imaging in rodent models, NVC has been widely used to assess the degree to which the neurovascular response is coupled or uncoupled in diseased and injured brains. We discuss three stimulation paradigms used to assess NVC in rodents (1) hindpaw and forepaw stimulation, (2) whisker stimulation, and (3) visual stimulation. Although auditory stimulation¹⁰⁹ has been performed to assess NVC, we do not discuss it in this review.

With electrical stimulation of the hindpaw or forepaw, several parameters can be modified to assess NVC. Modifiable parameters include the stimulation amplitude (current), pulse width, total stimulation duration, and frequency. The current delivered to the rodent paw ranges from 0.5 to 1.5 mA.²⁶^,¹¹⁰^–¹¹³ A higher current will typically lead to a larger hemodynamic response. However, care must be taken to not exceed a current that causes pain, as this response differs from a traditional NVC response. Current above 1.5 mA may elicit a pain response and should not be used to study NVC dynamics.¹¹⁴ Other modifiable parameters also have variability. The pulse width in studies ranged from $300 μ s$ to 1 ms, whereas the stimulation frequency can range from 1 to 10 Hz. Finally, the stimulation duration can vary from 2 to 60 s. Although these ranges are fairly large, the most common parameter set is an $\sim 1 mA$ current, a pulse width of $300 μ s$ , a stimulation frequency of 3 Hz, and a stimulation duration of 2 s.²⁶^,¹¹¹^,¹¹³ One disadvantage to electrically stimulating the paw of rodents is that the animal needs to be under anesthesia. Studies have used a large variety of different anesthetics, such as isoflurane,¹¹⁵ alpha-chloralose,²⁶ dexmedetomidine,¹¹⁶ medetomidine,¹¹⁷ and ketamine-xylazine.¹¹² The type of anesthetic alters the NVC response to stimulation, so care should be taken when selecting an anesthetic in NVC studies.

Whisker stimulation is a standard method to assess NVC in rodents. Several different methods can be used to stimulate whiskers, such as using an air puff,¹¹⁸^,¹¹⁹ a stepper motor with an attachment,¹²⁰^,¹²¹ or electrically stimulating the trigeminal nerve or whisker pad.¹²²^,¹²³ Furthermore, a single whisker¹²⁴^,¹²⁵ or multiple whiskers¹²⁶ can also be used to assess NVC. Electrically stimulating the trigeminal nerve is similar to electrically stimulating the hindpaw and forepaw. Studies that have used electrical stimulation had a wide range of parameters. The parameters had the following ranges: current delivered was from 0.2 to 1.5 mA, pulse width was from 0.1 to 1 ms, stimulation frequency was from 0.25 to 40 Hz, and a stimulus duration from 4 to 30 s. As opposed to paw and whisker electrical stimulations, air puff and stepper motor methods are advantageous since imaging can be performed in awake rodents. Studies that used a stepper motor have an object attached to the motor to deflect the whisker in a repeatable manner. Stepper motor experiments modified the stimulation frequency and duration. Most studies use a frequency between 3 and 10 Hz with a duration between 2 and 20 s. Finally, the air puff modified the input settings for stimulus duration from 1 to 30 s, frequency from 3 to 5 Hz, and pulse duration from 1 to 100 ms.

Visual stimulation is advantageous such as whisker stimulation since anesthesia is not required to assess NVC. Many different methods exist to elicit an NVC response to visual stimulation. It is generally performed using a monitor that displays a spatial pattern of rotating bars on the screen. These spatial patterns can have different gray levels, degrees to how much they are rotated, different spatial frequencies, and display duration. The display of spatial patterns elicits an NVC response in the visual cortex, instead of the somatosensory cortex with other NVC methods.¹²⁷^–¹²⁹

There are different analysis methods to quantify the NVC response. Two standard quantifiable metrics are the magnitude and rate of change in response to stimulation. The magnitude can be quantified as the peak response or the maximum change in CBF [Fig. 2(b)] or HbT. Furthermore, the rate of change is related to how fast the hemodynamic response is to stimulation, which can be calculated by taking the slope of the temporal hemodynamic data [Fig. 2(b)]. Using these quantifiable metrics, statistical tests can be performed comparing diseased animals with control animals or among the same animals over time.

4.3.

Cerebrovascular Reactivity

CVR is the ability of cerebral vessels to dilate in response to vasoactive stimuli. CVR is widely used in humans to assess cerebrovascular health and function, such as in patients with cerebral amyloid angiopathy (CAA).¹³⁰^,¹³¹ With widefield functional optical imaging in rodents, there are two main approaches to assess CVR: (1) injection of acetazolamide or (2) gas inhalation. These approaches do not have as many knobs to turn as NVC approaches for the experimental setup. Acetazolamide increases ${pCO}_{2}$ and acts as a potent vasodilator, which enables the study of CVR. Studies that use acetazolamide administer it intravenously with a concentration of 14 to $50 mg / kg$ .¹³²^,¹³³ Acetazolamide has also been given intraperitoneal in awake mice with a concentration of $90 mg / kg$ .¹³⁴ Since acetazolamide has a relatively long half-life, only the rise-time and maximum change in hemodynamics can be analyzed. Gas inhalation methods use ${CO}_{2}$ to induce hypercapnia, which induces vasodilation. These experiments typically have a baseline period where the gas inhaled is under room air with isoflurane concentrations between 0.75% and 2%. After the baseline period, a period where the ${CO}_{2}$ concentration is increased to a level between 5% and 10%, with the balance room air. The time under increased ${CO}_{2}$ concentration ranges from 1 to 10 min.¹³⁵^–¹³⁸ After this period, conditions are changed back to baseline conditions. The gas inhalation is advantageous compared with using acetazolamide because one can quantify what happens during the vasodilatory period of increased ${CO}_{2}$ concentration and after the vasoactive stimulus is removed.

Similar to quantifying NVC dynamics, there are different methods to assess CVR. The magnitude can be described as the peak response or maximum change in CBF [Fig. 2(b)] or HbT. Furthermore, a rate of change can be calculated by taking the slope of the temporal hemodynamic data after the onset of the vasoactive stimuli. In addition, a rate of change can be analyzed as the time when the hemodynamic response reaches 50% of the peak response ( $t_{50 rise}$ ) or when the vasoactive stimulus is removed and the hemodynamic response decays, a $t_{50 fall}$ can be obtained. Finally, the area under the curve (AUC) from the temporal hemodynamic data is a common metric that can be calculated during the hypercapnia period. These metrics allow for statistical tests to be performed that can compare diseased animals to control animals or among the same animals over time.

5. Applications