2.1.

Laminar Optical Tomography

In a typical LOT system, a noncontact laser scanning setup^5,10 similar to confocal microscopy is used to raster scan the tissue by a focused beam of light directed by a pair of moving galvanometer mirrors. The returning light is detected from both the focal point of the scanning beam and at a series of increasing distances (0.0 to 1.6 mm) from the beam’s focal point, allowing the superficial and deeper structures to be distinguished for 3-D image reconstruction, see Fig. 1.

Fig. 1

FLOT measurement geometry and source–detector configuration in a line pattern illumination. The emitted fluorescent signal is collected from up to nine detector arrays parallel to the line illumination with separations of 0.0 to 1.6 mm. Seven source–detector pairs with the separation between 0.0 to 1.2 mm are shown.

In the fluorescence LOT scanner, the excitation light can be delivered in the form of scanning point source⁵ or sweeping line illumination.¹³ In point illumination, to acquire high-resolution data, fast photodetectors are required to reduce the scan time. The method of choice in this project was the line illumination and to reduce the scan time, an electron multiplication CCD camera was used to acquire data along up to nine detector arrays (0.0 to 1.6 mm) parallel to the illumination pattern.

2.2.

Experimental Setup

The schematic of the implemented FLOT system is shown in Fig. 2. A single-mode fiber-coupled blue laser (LP450-SF15, $450\text{-}\mathrm{nm}$ single-mode pigtailed laser diode, Thorlabs Inc., Newton, New Jersey) is used to achieve a sharp focus when delivering the excitation light. The light from the blue laser is collimated (F810SMA-635, Thorlabs Inc., Newton, New Jersey) and then passed through a cylindrical lens (LJ1695RM, Thorlabs Inc., Newton, New Jersey) to obtain the line-field illumination with a full line-width at the half maximum of $42\mu \mathrm{m}$ on the sample. Next, the beam is passed through a dichroic mirror (FF440/520-Di01, Semrock, Rochester, New York). Wavelengths shorter than a threshold (excitation light) pass through the dichroic mirror, whereas longer wavelengths (emission light) are reflected toward the imaging components of the system. After passing through the dichroic mirror and before reaching the scan lens, an achromatic doublet lens (AC254-060-A-ML, Thorlabs Inc., Newton, New Jersey) with a focal length of $f_2=60\mathrm{mm}$ is placed in a way that the distance between the cylindrical and achromatic lens is $f_1+f_2$. Then the excitation light hits on a pair of moving $X$ and $Y$ galvanometer mirrors (GVS212, Thorlabs Inc., Newton, New Jersey). The galvanometer mirrors are controlled by the computer software prepared under LabVIEW (National Instruments, Austin, Texas) to steer the collimated beam through a scan lens. The scan lens (LSM03-VIS, Thorlabs Inc., Newton, Austin) focuses the line beam on the surface of the tissue to scan the predefined FOV by converting the angular changes on one of the galvanometer mirrors into a horizontal translation of the scanning line.

Fig. 2

(a) Block diagram of the FLOT system. Collimated laser light passes through a cylindrical lens for line illumination. A dichroic mirror separates the illumination from the imaging path. Excitation light is focused on the sample by a scan-lens. A pair of galvanometers is used to scan and descan the tissue. The emitted light is reflected by the dichroic beam splitter and passes through a narrow-band emission optical filter and then imaged on the EMCCD sensor. (b) Snapshots of the developed FLOT system.

The emitted light produced by the fluorophores inside the tissue is collected by the same scan lens and descanned by the galvanometer mirrors. To this point, the excitation and detection paths overlap. Nonetheless, when the returning light is propagated backward in the system and reaches the dichroic mirror, only the emission wavelengths are reflected toward the light detection components of the system. To completely remove the unwanted wavelengths, a narrow-band green emission filter is used in front of the detector. The light entering the optical filter is a converging beam. Therefore, to eliminate problems associated with placing an optical filter in front of a focusing beam, such as the leakage from the emission filter and change in the focal plane caused by the thickness of the filter glass, a two-piece collimated emission-port adapter (XT2, Photometrics, Tucson, Arizona) is used to provide a collimated (infinity corrected) space for a filter wheel. Finally, a highly sensitive electron multiplication charge coupling device (EMCCD) (Evolve 512, Photometrics, Tucson, Arizona) is used to acquire data along the lines parallel to the illumination pattern at each step. Therefore, in this configuration, photons that travel through various depths are detected simultaneously in each single frame captured by the camera.

2.3.

Forward Model for Light Propagation

Design and implementation of any optical tomography system requires a mathematical forward model, describing light propagation in the medium, and an inverse problem solver to reconstruct unknown parameters by fitting the measurements to the outcome of the forward model. Therefore, modeling and simulation of photon propagation in the medium is an important first step toward any optical device design.

Propagation of photons inside biological tissue can be modeled by the radiative transport equation (RTE). However, the RTE is difficult to solve numerically and exact solutions only exist for relatively simple geometries. The RTE is simplified by applying a set of approximations to take the form of a diffusion equation. In the diffusion approximation regime, it is assumed that the radiation in the medium is almost isotropic; therefore, this model is not suitable for small source–detector separation setups like the proposed FLOT scanner, where the distance between a source and a detector can be less than the photon scattering length. In FLOT setups, Monte Carlo (MC) simulations¹⁴ are employed to generate the sensitivity matrix for each source–detector pair.⁵ A typical simulated sensitivity matrix for multiple source–detector separations in a phantom with optical properties of absorption coefficient ${\mu }_{\mathrm{a}}=1{\mathrm{mm}}^{-1}$, scattering coefficient ${\mu }_{\mathrm{s}}=9{\mathrm{mm}}^{-1}$, and anisotropic factor $g=0.9$ is shown in Fig. 3.

Fig. 3

Simulated sensitivity matrix using MC light propagation model¹⁴ for six different source–detector separations in a scattering medium with the optical properties: ${\mu }_{\mathrm{a}}=1{\mathrm{mm}}^{-1}$, ${\mu }_{\mathrm{s}}=9{\mathrm{mm}}^{-1}$, and $g=0.9$.

2.4.

Inverse Problem

The 3-D distribution of the fluorescent concentration in samples was determined by solving the following equation:

One shortcoming in most optical tomography systems is that the sensitivity of the measurements drops exponentially as a function of the photon penetration depth. The result of lower sensitivity at larger depths is the inaccuracy in depth localization where the reconstruction methods are biased to move the objects toward the superficial areas.^19,20 In Ref. 20, a specific depth compensation algorithm is proposed, which operates to counterbalance this exponential decay of light penetration by modifying the elements of the weight matrix $W$. To apply this compensation algorithm, we assume that the image to be reconstructed is made from $l$ separate layers. In this case, the sensitivity matrix $W$ can be decomposed to $l$ submatrices

2.5.

Phantom Fabrication and Calibration Procedures

To evaluate the performance of the system, we needed a phantom with the optical properties close to brain tissue. Based on our previous experiments,²¹ for rat cortical tissue, typical optical properties are ${\mu }_{\mathrm{a}}=0.05{\mathrm{mm}}^{-1}$, ${\mu }_{\mathrm{s}}=11{\mathrm{mm}}^{-1}$, and $g=0.9$ at 450-nm wavelength. We decided to make silicone-based microfluidic tissue phantoms, which are known to be robust, cost effective samples that have a long shelf life. Using microfluidic fabrication technology, we embedded a microchannel structure inside the phantom, which was then filled with fluorescence solutions. Since the channel was embedded within the substrate, there was no need to add extra components to hold the channel.

We fabricated our phantoms in pure polydimethylsiloxane (PDMS) and its associated curing agent (Sylgard 184 silicone elastomer kit, Ellsworth Adhesives, Germantown, Wisconsin) following a 10:1 mixing ratio, to obtain a solid yet flexible structure. Titanium dioxide (${\mathrm{TiO}}_2$, Sigma Aldrich, St. Louis, Missouri) and India ink (Chartpak Inc., Leeds, Massachusetts) were added to this substrate to introduce appropriate amounts of scattering and absorption, respectively. The first step in PDMS molding is the mold design. Such molds come in a variety of shapes and sizes and are designed and fabricated using the soft lithographic method. The molds we used had rectangular channels with different widths and heights. The master for replica molding of the PDMS structure was fabricated via photolithography on a silicon wafer using negative photoresist (SU-8, Microchem Co., Newton, Massachusetts).²²

Details of the microfabrication process are discussed in Refs. 22 and 23. The surface conditioning of the mold is an important factor in preventing PDMS from sticking. After preparing the mixture, it was poured into the molds and placed into a vacuum chamber for at least 30 min to remove entrapped bubbles. Next, the prepared mold was placed on a level surface in a 55°C heated chamber (oven) for 8 h. After demolding the phantom and punching the holes for the inputs/outputs of the channels, the PDMS structure was treated by a plasma generator (Electro-Technic Products Inc., Chicago, Illinois) and attached to another PDMS layer for permanent bonding without the use of any intermediary contact gel or air gaps. Figure 4 shows the final phantom and a microscopic image of the buried microchannel in PDMS substrate.

Fig. 4

(a) Cross section of the microfluidic phantom, a channel with the size of ($200\mu \mathrm{m}\times 150\mu \mathrm{m}$) is embedded at the depth of 1 mm, (b) and (c) final phantom and its image under the microscope. The channel was filled with FAD solution as fluorescent liquid using a syringe pump.

Optical properties of the prepared phantom were measured using a double-integrating sphere setup.^24,25 The schematic of the setup is shown in Fig. 5. The main variables which are measured through these experiments are the diffused reflectance ($M_R$) and transmittance ($M_T$), defined by the amount of light reflected by and transmitted through the sample, normalized to the intensity of the incoming light. We used the following equations to calculate these variables:²⁴

Fig. 5

(a) Schematic of the double-integrating sphere setup used to measure the optical properties of the microfabricated tissue phantoms. (b) A snapshot from the double-integrating sphere optical setup.

2.6.

Surgical Procedures for In Vivo Experiments

We used rats to run a sequence of in vivo experiments. All procedures were carried out in the facility accredited by the Association for Assessment and Accreditation and Laboratory Animal Care and approved by the University of Wisconsin–Milwaukee Institutional Animal Care and Use Committee and conducted within the ethical guidelines of the National Institutes of Health.

One to four weeks before the experiments, animals were anesthetized with isoflurane in 100% oxygen (induction occurred with 4% isoflurane and maintained with 2%). During experiments, the head of the animal was fixed in a stereotaxic apparatus. For viral transfection, purified AAV9-CAG-GFP vector was infused into the primary somatosensory cortex (forelimb) (A/P $-1.0\mathrm{mm}$, M/L $+3.8\mathrm{mm}$) and the primary visual cortex (A/P $-5.3\mathrm{mm}$, M/L $-3.8\mathrm{mm}$). The depth of injection was systematically varied to include 0.8, 1.0, 1.5, and 2.0 mm. The injection volume was also systematically varied to include 0.2, 0.5, and $0.75\mu \mathrm{l}$. The virus was injected at the rate of $0.05\hspace{0.17em}\mu \mathrm{l}/\min$ by using a $10\mu \mathrm{l}$ syringe and a 34-gauge needle mounted on the stereotaxic automated injector. The injector was left in place for 10 min following the injection to allow the diffusion of the virus away from the injector. 1 to 4 weeks later, animals were ready to perform the experiments.

Immediately following the completion of experiments, animals were deeply anesthetized with isoflurane, transcardially perfused with 0.2 MPBS followed by 10% buffered formalin. Brains were removed and postfixed in 10% formalin for 24-h and then transferred to 30% sucrose/PB. The brains were then frozen, sectioned into coronal slices of $200\text{-}\mu \mathrm{m}$ thickness, then mounted on glass slides, and coverslipped with UltraCruz mounting medium (Santa Cruz, California) containing $1.5\mu \mathrm{g}/\mathrm{ml}$ DAPI (for nuclear counter staining).

3.1.

Phantom Results

After fabricating tissue phantoms, flavin adenine dinucleotide (FAD) fluorescent solution was injected into the channel using a syringe pump (Genie Plus, Kent Scientific Corporation, Torrington, Connecticut). Then an area of $3\mathrm{mm}\times 3\mathrm{mm}$ was scanned by a line-shaped laser beam at 450-nm wavelength, with a resolution of $70\mathrm{lp}/\mathrm{mm}$. During each illumination, one image was captured by the EMCCD camera and transferred to a computer for further processing and extracting the information of different detectors. A total measured dataset of $200\times 200\times 7\text{voxels}$ was acquired by the system and nearest-neighbor interpolation method was employed to down-sample it to $60\times 60\times 7\text{voxels}$, which was used as the input for our reconstruction algorithm.

As an example, for a microchannel buried at a depth of $800\mu \mathrm{m}$, the experimental raw data obtained by detectors number 1 to 6 is shown in Fig. 6. This raw data were fed into our reconstruction algorithm to generate the image of fluorescent distribution. The reconstruction was performed in 2-D planes and then a 3-D structure of the fluorescent object was obtained by stacking all the 2-D images together.¹³ The reconstruction process of one 2-D cross section, after 3000 SART iterations, takes $\sim 20\mathrm{s}$, when the program is run by a 64-bit operating system, Windows 7 home premium, 8GB RAM, Intel core(TM)i3 CPU at 3.3 GHz.

Fig. 6

Experimental raw data obtained by the detectors number 1 to 6 for a microchannel phantom with a rectangular cross section ($200\mu \mathrm{m}\times 25\mu \mathrm{m}$) embedded at the depth of $800\mu \mathrm{m}$.

To find the exact depth of the channel and confirm the reconstruction results, we scanned our sample by a spectral-domain optical coherence tomography (SD-OCT) system, which was developed and tested in the lab prior to current experiments.^27,28 As displayed in Fig. 7, the OCT images (left panel) show the actual depth of two microchannels buried at depths of 800 and $1200\mu \mathrm{m}$, and the right panels show the fluorescent reconstruction results generated by the FLOT system. By increasing the depth of the microchannels, the point spread function of the system, and as a result, the diameter of the reconstructed channel increase.

Fig. 7

The reconstruction results of two phantoms, one with the channel size of ($200\mu \mathrm{m}\times 25\mu \mathrm{m}$) at the depth of $800\mu \mathrm{m}$, and one with the channel size of ($150\mu \mathrm{m}\times 25\mu \mathrm{m}$) at the depth of 1.2 mm. The exact depth of the microchannel was measured by the OCT system and is shown on the left panel for comparison. A 2-D ZX cross section of the reconstructed channel by SART algorithm is shown on the right panel.

The full width at half maximum of the reconstructed microchannel along the $x$- and $y$-dimensions are 250 and $450\mu \mathrm{m}$ for the channel at the depth of 800 $\mu \mathrm{m}$, and 450 and $750\mu \mathrm{m}$ for the channel at the $1200\text{-}\mu \mathrm{m}$ depth. Figure 8 shows the 3-D reconstructed image of the $1200\text{-}\mu \mathrm{m}$ depth microchannel filled with the FAD solution.

Fig. 8

The 3-D reconstructed image of a microchannel at the depth of $1200\mu \mathrm{m}$ filled with FAD solution.

3.2.

In Vivo Results

All in vivo experiments were conducted on thinned-skull rats (Fig. 9). Each animal received multiple injection of different volumes to express green fluorescent protein (GFP) in cortical areas at different depths and expression levels. Two different mechanisms were employed to compare the FLOT reconstructed results with the actual GFP distribution. After scanning the tissue and collecting the required measurements for the fluorescent tomography, we used our single optical fiber fluorescence probe system²⁹ to clarify the axial distribution of the GFP inside the brain around the injection site. This system uses a thin multimode optical fiber, which serves as the head of the probe, to deliver blue laser pulses (473 nm) to the region of interest and guide a sample of the emission signal back to photodetectors. After scanning the tissue with the fiber probe system, the animal was sacrificed and the brain tissue was extracted, sliced, and imaged by a confocal fluorescence microscope to further clarify the real spatial distribution of fluorescence proteins.

Fig. 9

In vivo experiments were conducted through a thinned-skull rat brain transfected with GFP. The returning fluorescent light was captured by the successively defined detectors ($D_0$ to $D_6$) while the laser beam scans the tissue (scale bar is $500\mu \mathrm{m}$).

Figure 10 shows the experimental raw data collected from the right hemisphere of a rat brain which received injection at the depth of 0.8 mm. For this experiment, the reconstructed images are displayed in Fig. 11(A), where panel (a) shows the confocal image of a rat brain slice with GFP expression. A 2-D cross section of the reconstructed fluorescent distribution is shown in panel (b). Panel (c) shows the normalized axial distribution of GFP recorded by the single fiber probe system and the superimposed image is shown in panel (d) for comparison. Another set of in vivo experimental results is shown in Fig. 11(B), where the rat brain was injected on the left hemisphere at the depth of $800\mu \mathrm{m}$. In both experiments, the fluorescent reconstruction results are in good agreement with the data from the probe system and also with confocal images.

Fig. 10

Experimental raw data of an in vivo scan of a rat cortex through thinned skull, obtained by detectors number 1 to 9. The injection was performed in the right hemisphere, at the depth of $800\mu \mathrm{m}$, with GFP.

Fig. 11

Experimental results of in vivo scans of rat brains with the injection depth of (A) $800\mu \mathrm{m}$ in the right hemisphere, (B) $800\mu \mathrm{m}$ in the left hemisphere, and (C) $800\mu \mathrm{m}$ in the left hemisphere. In (A) and (B), the gene expression was successful: (a) a confocal microscopy image of a rat brain’s slice close to the site of injection, (b) FLOT reconstruction result, (c) curve displays fluorescence signal as a function of depth detected by the single optical fiber probe system, (d) superimposed image of the reconstruction result and the confocal microscopy image. In (C), the gene expression was not successful: (a) curve displays fluorescence signal as a function of depth detected by the single optical fiber probe system, (b) a confocal microscopy image of a rat brain slice close to the site of injection.

Despite the fact that all the animals were carefully injected at specific depths with predefined volumes, in a few cases, the gene delivery was not successful and almost no fluorescent signal was detected by the FLOT system. Figure 11(C) shows the experimental results from a case where almost no fluorescent signal was detected by the FLOT system. The data from the fiber probe system and the confocal microscopy images of the brain slices also confirmed that the gene delivery was not successful. These experimental results prove that such a tomography system provides valuable information regarding the depth and level of the expression of optogenetic proteins prior to the experiments without causing any damage to the brain tissue under test.