Visualization of evoked and spontaneous neuronal activity in vivo is of great importance for understanding brain functions.1,2 Localizing neural activities is a critical step in understanding the functional characteristics of neuronal interactions in the brain.23.–4 Due to both high spatial and temporal resolutions, optical imaging has become an important technique to investigate the neuronal and vascular responses to brain activation.184.108.40.206.–10 Based on either absorption, scattering, or fluorescence contrasts, various optical imaging methods have been demonstrated for functional brain mapping,4 including diffuse optical imaging (DOI),1112.–13 intrinsic optical signal (IOS) imaging,14,15 multiphoton microscopy (MPM),16,17 optical coherence tomography (OCT),1819.20.21.–22 and photoacoustic tomography (PAT).23 However, limited resolution (DOI), relatively low contrast (OCT), shallow penetration depth (IOS), or low data acquisition speed (MPM and PAT) have somehow limited their applications in imaging three-dimensional (3-D) neural activities, a process that needs both high temporal and spatial resolutions.24
Laminar optical tomography (LOT) was initially developed to image absorption contrast for hemodynamic changes.2526.27.–28 Soon after, it was rapidly adapted to fluorescent molecular imaging, applications termed either fluorescence laminar optical tomography (FLOT)3,28,29 or mesoscopic fluorescence molecular tomography (MFMT).30,31 Similar to diffuse optical tomography (DOT),32 LOT uses an array of photon detectors or a charge-coupled device (CCD) camera to collect photons emitted from locations at different distances away from the illumination position, enabling simultaneous detection of scattered light travelling through different depths in the tissue. LOT is sensitive to both absorption and fluorescence contrasts providing an axial resolution of to with several millimeters imaging depth.3,25,26,33 Recent studies demonstrated that angled illumination or detection modification (termed aFLOT) improves both resolution and depth sensitivity.3,34 Another rapid, noncontact depth-resolved imaging method called modulated imaging has also been developed for quantitative, wide-field characterization of optical absorption and scattering properties of turbid media, utilizing frequency-domain sampling and model-based analysis of the spatial modulation transfer function.3536.–37
In this review paper, we focus on applications in neuroimaging using LOT. We first introduce the physical principles and instrumentation of LOT that enable depth-resolved imaging. Then, we review representative applications in 3-D imaging of the hemodynamic changes and neural activities in vivo.
Laminar Optical Tomography Principle and Instrumentation
Laminar Optical Tomography Basic Principle
The working principle of LOT is based on light transport in tissues,38 which includes three primary physical processes: scattering, absorption, and fluorescence. The relative probability of occurrence for each process depends upon the type of sample imaged and the wavelength of light used.39 For in vivo brain imaging, scattering is the prevalent phenomenon. During light propagation, some of the photons will scatter out from the surface of the tissue. These photons will be captured by the detectors near the tissue surface with various separations from the light illumination/entrance position (source–detector separations). The light emerging at greater distances (i.e., larger source–detector separation) has a higher statistical probability of having travelled deeper into the tissue. By detecting the emerging light for a range of positions with different source–detector separations, it is possible to perform depth-resolved imaging of subsurface tissue structures through a proportional relationship between the source–detector separations and the average investigation depths.28 Figure 1(a) shows the cross-sectional diagram of a typical LOT source–detector configuration and representative photon paths. The detection geometry used in LOT is similar to the detection geometry used in DOT. In contrast to DOT, in which source–detector separations are typically several centimeters,24 LOT utilizes smaller source–detector separations (from several tens of microns to a few millimeters). As a result of this difference, information from a relatively shallow depth (millimeter or mesoscopic scale) is collected by the detectors, enabling tomographic imaging with a higher resolution compared to that of DOT.40
Mathematical expression of laminar optical tomography based on fluorescence contrast
By using “normalized Born,” it is very helpful to cancel the systematic errors related to measurements (e.g., ariations in gain between each detector).24,25,41 Also, in fluorescence LOT measurements, fluorescence signals detected from shallower layers are much stronger than those from the deeper perturbation signal. By normalizing to the original signal, it becomes possible to measure the change in fluorescence before and after the perturbation at different depths within a proper dynamic range.24
Mathematical expression of laminar optical tomography based on absorption contrast
For the absorption case, the detected photons scatter along a random walk path from the source to the detectors and usually the scattering is predominantly in the forward direction, so we assumed that each scattering event causes a relatively small change to the direction of the photon. The perturbed radiance can be expressed as3824 24
Forward Modeling in Laminar Optical Tomography
The spatial sensitivity matrix can be predicted using light propagation models, in which tissue scattering and absorbing properties, fluorophore characteristics (extinction coefficient and quantum yield), and the relative positions of the light source and detectors must be included. Several methods have been used to model photon propagation in scattering samples.42,43 Two common types of forward modeling methods are summarized below.
Radiative transfer equation based model
Time-dependent radiative transfer equation (RTE) can be used to accurately describe photon migration through tissue.44 The diffusion approximation (DA) to the RTE is often applied since solving RTE is complex and computationally expensive, and there are analytic solutions to the infinite and semi-infinite geometry with different boundary conditions of the diffusion equation. 45,46 DA has been the preferred forward model in DOT due to its computational efficiency and ease of implementation.40,42 It has been shown that DA cannot account for boundary measurements when there are structures within the scattering length from the surface of superficial tissues.47,48 In other words, DA is invalid and will not provide accurate results for the small source–detector separations typically used for LOT in mesoscopic regime.
An improved DA to radiative transport, delta-P1 approximation, has been developed and validated for LOT.49 While delta-P1 approximation can improve the computational efficiency, the estimation accuracy is moderate (with relative errors ). Also, there are larger relative errors when source–detector separations are very small or when the heterogeneity is very close to the source or detectors due to the intrinsic limitations of the diffusion model.49 However, for some applications in which high speed is needed, delta-P1 approximation can be a good choice, especially for LOT based on fluorescence contrast. Furthermore, by taking advantage of the delta-isotropic phase function, the phase function corrected DA approach improves the performance of the DA and agrees with experimental results to an excellent degree, opening a potential for modeling photon propagation in LOT.50
Monte Carlo methods
By tracing the random walk steps that each photon packet takes when it travels inside the tissue, the Monte Carlo (MC) method has been used for simulating light transport in tissues for over three decades. 51,52 MC provides a flexible and rigorous solution to the problem of light transport in turbid media with a complex structure.53,54 The MC method is able to solve the RTE with a desired accuracy; hence, it is well suited to simulating the light propagation in mesoscopic regimes.40,47,54,55 Compared to other analytical or empirical methods, large packets of photons need to be simulated ( to ) to obtain simulations with stochastic accuracy, requiring intensive computation and a great deal of time. A variety of methods for speeding up MC simulations, including scaling methods,56 perturbation methods,57 hybrid methods,58 variation reduction techniques,59 and parallel computation,60,61 has been developed.54 The majority of LOT/MFMT work employs the MC method to compute the optical forward problem.3,27,26,40,62,63 Most of the current LOT/MFMT results used only homogenous forward models for generation of the spatial sensitivity , while heterogeneities (such as large blood vessels and different tissue layers) could feasibly be considered into the forward model to achieve improved fitting between the data and the model.24,64 However, since the photon path becomes more uncertain as they scatter further, especially in complexed biological tissues, the mismatch between the mathematical model prediction and actual measurement may cause the degradation of resolution as a function of depth.3,24,28
Image Reconstruction for Laminar Optical Tomography
After knowing , the next step is to solve Eqs. (3) or (4) to estimate the 3-D distribution of the fluorescence (or absorption) changes . It is challenging to solve the inverse problem since the number of measurements available is often much smaller than the number of unknown 3-D image elements (voxels), which may result in a nonunique solution.65 Moreover, compared to transmittance geometry, solving the inverse problem for reflectance geometry is more challenging due to limited angular sampling.66 The typical approach to solve the inverse problems is to minimize using an iterative solver, with the estimate updated iteratively to minimize until a preset number of iterations is reached or when is below a set value (tolerance).40 The most common iterative methods employed in the field are the conjugate gradient method,31 the least squares method,67,68 and algebraic techniques.29,69 However, using only these iterative solvers is still not enough since the system is ill-conditioned and very sensitive to noise propagation.40 A regularization term is required to tolerate small mismatches between the actual measurement and the prediction using the forward model, which could result from systematic errors or noise and inaccuracies in the light propagation model.24 One typical formulation of the inverse problem in LOT/MFMT is24 It becomes the Tikhonov regularization if is the identity matrix, which has been commonly used in LOT/MFMT. 25,34 Large values of will generate superficially weighted smooth images with underestimated quantification.24 Radially variant has the potential to minimize high-frequency noise in the reconstructed image and produce constant image resolution and contrast across the image field,70 with being a diagonal matrix, whose elements are the square-root of the corresponding diagonal elements of ,32,71 and can be selected via -curve analysis.72
Laminar Optical Tomography Instrumentation
The first LOT system was reported in 2004.25 Since then, several improvements or alternative designs have been investigated.3,28,30,31,73,74 We first summarize two representative LOT systems in the following sections, and then LOT/MFMT with variant configurations and time-resolved protocol are also covered.
Laminar optical tomography based on absorption contrast
In the living brain, the main absorption contrast relies on the distinctive absorption spectra of oxy- and deoxyhemoglobin ( and HbR). At visible and near-infrared wavelengths, hemoglobin in blood is the most significant absorber. Changes in blood flow, blood volume, and oxygenation can be induced following functional activities in the brain.75 Changing the oxygenation status of blood can be inferred from the changes in the relative concentrations of and HbR. In this way, absorption properties of the brain can be related to the variations in the oxygenation state of hemoglobin and further connected to the localized neural activities. In brain research, the cortical hemodynamic response to stimulus provides a detectable signal, which can report the presence and location of neuronal activation.27,76,77 Furthermore, by using different wavelengths of excitation lights and the known absorption spectra of HbR and , the changes in the concentration of and HbR can be calculated. One of the main advantages of the method based on absorption contrast is that there is no need to use any extrinsic chemical probes, which makes this approach more feasible for clinical translation. The scattering of brain tissues is often assumed not to change significantly during the hemodynamic response.9 While this method can record the hemodynamic responses to functional activations in the brain, the inter-relation between neural activities and hemodynamic changes is still an area of intense research.78220.127.116.11.–83
The general schematic of a LOT system based on absorption contrast is shown in Fig. 1(b).27 Illumination light from one or more lasers is first coupled into a fiber and then collimated. The collimated light then passes through a polarizing beam splitter and is steered by a pair of galvanometer scanning mirrors onto a scan lens. Together with an objective lens, the beam is focused onto the surface of the tissue. Signals collected from the tissue will pass back up through the same light path and is descanned by the galvanometers. To be able to separate the incident light and the scatter light, a strongly polarized laser is normally used. Since specular reflections from optics and the tissue surface will maintain the same polarization, light that has been scattered several times will lose its original polarization quickly. About 50% of the scattered light emerging from the sample will be successfully separated from specular reflections through the polarizing beam splitter. That successfully separated light then passes through a lens and is focused onto a linear fiber bundle or directly on the detector array.27 Different from the pinhole setting in a confocal microscope, the line of fibers in the LOT system acts like seven axially offset pinholes, each leading to a separate detector corresponding to signals from different depths. As the focused beam is scanned over the surface of the tissue, scattered light emerging from the tissue at seven different depths will be collected simultaneously by the seven avalanche photodiode detectors, which results in seven two-dimensional (2-D) images per raster scan [e.g., if we choose over a field of view (FoV), then we have source positions and measurements from all the detector positions.]. A faster and more sensitive second generation LOT system that utilizes three excitation wavelengths and incorporates simultaneous imaging of both absorption and fluorescence contrast has also been reported.28
Laminar optical tomography based on fluorescence contrast
Exogenous fluorescent dyes, as well as transgenic methods, can also provide highly specific optical contrast enhancement. They can report functional parameters, such as changes in membrane potential [voltage-sensitive dyes (VSD)] or ion concentrations (pH-, calcium-, chloride-, potassium-sensitive dyes).4,84 By binding to the neural membrane and converting changes in transmembrane voltage into the fluorescence signal of the emitted light, VSD imaging provides a useful and direct manner in revealing activities of the neural networks in the brain with relatively high spatial and temporal resolution (up to microseconds).4 Compared to hemodynamics imaging based on absorption contrast (slow signal changes second time scale), fluorescence signal change originates directly from rapid changes of neural activities, which is at the millisecond scale.4
The main difference of the LOT system based on fluorescence contrast compared to the LOT based on absorption contrast is the use of a dichroic mirror (for fluorescence) or a polarizing beam splitter (for absorption), and a long-pass or band-pass optical filter (depending on the type of fluorophores) before the detectors to further block the excitation light. Figure 2(a) shows one schematic diagram of a recently developed line-scanning aFLOT system.3 The excitation light from a 637-nm laser diode is collimated and then coupled into a fiber to shape the light beam. An objective lens is then used to collimate the light from the fiber and the collimated light is further expanded into line-field illumination using a cylindrical lens. The emitted fluorescent light is collected through the objective lens, dichroic mirror, emission filter, and finally focused onto a high-speed CCD camera. Since the illumination light path is not confocal with the emission light path, the dichroic mirror is optional. Both fluorescence and reflectance images can be recorded by changing the corresponding filters. This system achieves scanning by moving the sample, using a precisely controlled motor stage laterally in the scanning direction (perpendicular to the line illumination direction).3 The acquired image is in XYS format (where and represent and dimensions of the 2-D image acquired by CCD, represents the different scanning positions, , ; , with a pixel size of in the specific system introduced). To reconstruct the images, the first-order Born approximation between the measurement and the fluorophore distribution was applied on the cross-section , with 100 source–detector pairs and 100 scanning positions chosen to constitute 10,000 measurement modes.3 And each reconstructed including with a pixel size of . Weight matrix is therefore of size .3 Finally, the 3-D was constituted by superimposing individual in the direction (illumination line direction).3 Although a graphics processing unit was applied for parallel computation, the reconstruction process for one 3-D image still takes around 10 min. However, with parallel computation developing so rapidly, we definitely believe that the time needed for MC modeling and image reconstruction will not be a concern in the future.
Laminar optical tomography/mesoscopic fluorescence molecular tomography with variant configurations
Several LOT/MFMT systems have been reported using different source/detector and scanning configurations, including point scanning with sparse array detection,26 point-scanning with dense array detection,68 and line-scanning with dense array detection.29,34 Laser diodes or solid-state lasers are typically chosen as the light source.40 A point-focused light illumination with a one-dimensional (1-D) array detector is usually used in the LOT/MFMT system.26 The 3-D images are obtained through 2-D raster scanning of the illumination point, as shown in Fig. 1(b). Alternatively, a line-field illumination with a 2-D array detector [CCD or electron multiplying CCD (EMCCD), or 2-D photomultiplier tube] can be utilized, and then only 1-D scanning is required to acquire the 3-D information, as shown in Fig. 2(a). Line-scan imaging can alleviate the complicated 2-D scanner design. While similar to confocal microscopy, the line illumination mode potentially decreases the axial resolution.85
We should note that in LOT/MFMT, a detector closer to the illumination source will receive high light intensity, while a detector with larger source–detector separation collects a much lower signal, meaning that the detector with a larger source–detector separation may need higher gain. If possible, each detector’s gain should be chosen to match its optimal signal intensity. However, when using CCD/EMCCD as the array detector, all pixels have the same gain and exposure time, which is a main drawback of such a design and may limit the system’s dynamic range.
Laminar optical tomography/mesoscopic fluorescence molecular tomography recording with time-resolved protocol
Hemodynamic response is usually slow and happens in seconds; therefore, the LOT system with imaging speed can be utilized in studying hemodynamic responses.27 In order to use LOT/MFMT to record VSD dynamics, which reflects the cellular processes at the millisecond scale,4 a new acquisition protocol has been investigated.3,26 Hillman et al.26 first reported a time-resolved FLOT system with an effective frame rate of 667 Hz imaging speed on perfused rat hearts. Basically, it requires the biological response to be repeatable for each stimulation trial and records the time relation of the acquired data with respect to the stimulus. Recently, Tang et al.3 used aFLOT to visualize neural activities in the living animal brain using VSD. One representative image acquisition protocols for time-resolved aFLOT is illustrated in Fig. 2(b). The line-illumination light is focused on the border of the desired FoV at the beginning of recording. An experimental session including all the time related images is acquired at each scanning position (e.g., for the whisker stimulation experiment, the experimental session consists of the recording before stimulus onset, and long-enough recording after stimulus onset to cover both the activation and recovery of neuronal signals). Then, another experimental session is performed to obtain the dataset at the next illumination/collection area by moving a step in the scanning direction. This process is repeated until the entire FoV is covered. The process is similar to previously published point-scanning FLOT.26 For reconstruction of data recorded using the time-resolved protocol, the measurements from multiple detectors at different scanning positions with the same frame number (or corresponding to the same time points in a dynamic process) should be first regrouped as one dataset. Then, the entire time course of 3-D neural responses can be obtained by repeating the reconstruction process for all time points.
Representative Applications in Neuroscience
In this section, we will review three representative examples of LOT/MFMT in vivo imaging applications in neuroscience. The first two are imaging hemodynamic responses using absorption contrast, targeting the cortex and the spinal cord, respectively. The last example is focalized on imaging the neural activities evoked in the whisker system of mice by deflection of a single whisker in vivo based on fluorescence contrast using VSDs.
Laminar Optical Tomography for In Vivo Imaging of Hemodynamics in Cortex
The cortical hemodynamic response to stimulus can provide a detectable signal, which can indicate the presence and location of neuronal activation.27,77 The LOT system shown in Fig. 1(b) has been used to perform depth-resolved optical imaging of hemodynamic responses to forepaw stimulation in the exposed rat cortex in vivo.27 This new system utilizes multiwavelength scanning (473 and 532 nm) and acquires up to 10 3-D images per second.27 Figure 3(a) shows the CCD camera image of the FoV, and the LOT FoV is identified by the dotted white square. Figure 3(b) shows the HbR, , and total hemoglobin (HbT) functional changes 0.6 s after cessation of the stimulus from the horizontal LOT slices. Different slices can sample and correspond to different depth regions (e.g., the slice at corresponds to deeper cortical layers (layers III–IV) and predominantly samples changes in the capillary bed). The advantage of LOT can be revealed more clearly in Figs. 3(c) and 3(d).27 Vertical slices from the 3-D data in the – plane transecting a surface draining vein and the focal capillary region are extracted, which are labeled (i) and (ii) in (a) and (b), respectively.27 Based on the distinctive differences among the dynamic behaviors of the arteriolar, capillary, and venous responses observed, LOT responses can be further segmented into regions, as shown in Fig. 3(e). Then, the distinctive functional temporal signatures associated with the hemodynamic response in different vascular compartments can be further investigated.27 Rapid, full-field two-photon microscopy was also applied to validate the LOT findings and to further explore the vascular mechanisms of the hemodynamic response in vivo.27
Hemodynamics of Spinal Cord
Spinal cord trauma is a serious injury and it is crucial to understand post-traumatic neuronal reorganization in the spinal cord.86 In a recent study, Ouakli et al.86 investigated LOT for imaging spinal cord activation following electrode stimulation in the left hind paw, with simultaneous intrinsic optical imaging of the cortex in rats. This study indicates that LOT is a potentially powerful imaging method to study activation in the spinal cord and subsequent disruption after injury. The LOT system used in this study is similar to the system illustrated in Fig. 1(b) except only a 690-nm excitation light source was used, which means HbR was the principal chromophore providing absorption contrast.86 Figures 4(a) and 4(b) show the simultaneously quantified activation in the cortex and spinal cord, in which Fig. 4(a) represents LOT signals from the first detector in the spinal cord at different time points after stimulation. An initial slight dip can be seen locally in the dorsal vein at 3.6 s, which may provide evidence that neurons initially use up local oxygen more quickly than it is being supplied by the gradually increased blood flow.9 At 10.80 s, the increase appears on the left side of the cord indicating blood is finally drained through the dorsal vein after electrode stimulation in the left hind paw. One interesting observation is that the hemodynamic response in the spinal cord is ipsilateral compared to the contralateral response in the cortex, as shown in Fig. 4(b). Figures 4(c) and 4(d) show the reconstructed images. Most response is ipsilateral on superficial layers of the dorsal horn, as shown in Fig. 4(c), which corresponds well to the location of gray matter in the spinal cord in Fig. 4(d). On the other hand, the signal in deeper sections of the spinal cord appears more diffused. Also, contralateral response is observed, which may originate from interneuron connections. Further improvements are needed to increase the imaging depth in order to image deeper signals.
Fluorescence Laminar Optical Tomography for Brain Cortex Imaging with Voltage-Sensitive Dyes
The rodent whisker-barrel system serves as an excellent model to investigate the development, organization, function, and plasticity of mammalian sensory pathways.87 In this system, whiskers on the snout are represented by neuronal modules at the brainstem (barrelettes), thalamic (barreloids), and neocortical (barrels) levels.8788.–89 The barrel fields are close to the brain surface, therefore, many imaging studies have been carried out at this region.10,9091.92.93.94.95.96.97.–98 However, virtually all of the previous in vivo VSD cortical imaging studies obtain only 2-D depth-integrated responses, lacking depth-dependent information. Recently, by combining aFLOT with VSDs, in vivo 3-D neural activities in the whisker barrel cortex of mice following deflection of a single whisker have been demonstrated with 5-ms temporal resolution. This study utilized the aFLOT system and time-resolved protocol, as shown in Figs. 2(a) and 2(b). In this study, system performance was first characterized by imaging a quantum-dot-embedded hydrogel and glass capillary tube in living mouse brain. Although the accuracy and resolution of aFLOT degraded with depth, the reconstruction distortion was within up to depth (see Fig. 5), covering up to layer V in the cortex of young mice. In addition to the 3-D static system performance evaluation, a rapid response to direct electrical stimulation through the use of a bipolar electrode in the living mouse brain was observed to verify the time-resolved protocol. Then, aFLOT was used to record 3-D neural activities evoked in the barrel cortex by deflection of a single whisker in vivo. The C2 whisker was used for 20-ms mechanical stimulation, performed by an air-puff stimulator. Figure 6 shows the 3-D aFLOT reconstructed changes in fluorescence [ (%), ordinate] following C2 whisker stimulation. The appearance of the signal was about 30 ms after the stimulus. After reaching its peak at 50 ms, the activated signals disappeared gradually. Furthermore, the results can be used to extract the 3-D dynamic activity patterns, which serve as a 3-D imaging tool to study the neural network processing in different parts of the cerebral cortex.3
The 3-D imaging techniques enabling examination of spatiotemporal patterns of neuronal activities will definitely provide more details in explaining how neural networks operate in the living brain. In this paper, we first review the fundamental basis of the LOT/MFMT system, including the MC method, which is usually used to model photon propagation, and common iterative methods and regularization models for image reconstruction. Specific LOT/MFMT systems based on absorption and fluorescence are described in detail, including the data acquisition process, data size, and image reconstruction process. And, finally, three representative examples of LOT/MFMT in vivo imaging applications in neuroscience are presented.
By adopting a microscopy-based setup and DOT imaging principles, LOT/MFMT can perform 3-D imaging with higher resolution 100 to than DOT and deeper penetration (2 to 3 mm) than confocal and two-photon microscopy while there is always a trade-off between axial resolution and penetration depth. Due to the limited penetration depth of LOT/MFMT, thinning or exposing the skull is usually necessary when imaging the cortical hemodynamic responses or neural activities using LOT/MFMT, which indicates that imaging the brain with LOT/MFMT is mainly applicable in small-sized animals.3,27 In such cases, LOT/MFMT has been focused to applications in which tissues of interest are superficial, such as the exposed mouse brain and skin, as well as oncological applications.24,74,34,99 For instance, LOT/MFMT has been applied to image skin cancers to describe the depth and thickness of pigmented skin lesions in clinical settings99 and has also been employed to image the biodistribution of a photodynamic therapeutic agent with ultrasound co-registration in skin cancer models in vivo, though we can see there is still a long way to human clinical translation.74 Imaging of internal organs using LOT/MFMT can be potentially achieved via endoscopic, intraluminal, or an intrasurgical imaging setup.24 Laser scanning microscopy (e.g., confocal and two-photon microscopy) aims to reject light that has been scattered to obtain high-resolution images of the tissue either by isolating the signal from the focus using a conjugated pinhole or by employing the nonlinear effect.100,101 Instead, LOT/MFMT takes advantage of the scattered light so that they are much more sensitive to the optical signal changes in the tissue. LOT/MFMT obtains depth-resolved information by measuring the scattered light emerging from the tissue using detectors at different distances from the source illumination position, instead of scanning the tissue in the axial direction, which can dramatically improve the data acquisition efficiency. On the other hand, since the detected light undergoes multiple scattering in the tissue, the resolution of LOT/MFMT cannot compete with laser scanning microscopy. Moreover, estimation of photon migration using the mathematical models could not be exact, especially for the complicated biological tissues and the path of photons that become more uncertain as they scatter further. LOT/MFMT faces resolution deterioration of these reconstructed images as a function of depth.3,24,28 A combination of dense spatial datasets with regularization terms like compressive sensing-based methods has the potential to push LOT/MFMT resolution close to or beyond, even at depths of several millimeters.40 In terms of image visualization, since laser scanning microscopy uses a more “direct” way to obtain intensity of every pixel in the image, it can achieve nearly real-time image feedback. While applying MC modeling and regularization term to solve the inverse problem, LOT/MFMT has a high computational burden and is time-consuming especially for a system with high source–detector density, as mentioned in Sec. 2.4. As a result, for now, LOT/MFMT cannot provide the reconstructed image in real time, perhaps restricting the translation to clinical applications. With the advent of the supercomputer, the time needed for high-burden computation in LOT/MFMT could be alleviated significantly.
With the advantages of overcoming the scattering limit, offering relatively high resolution, multiplexing capabilities, large FoV, and high acquisition speed, LOT provides a promising imaging technique to investigate 3-D neural activities in a minimally invasive manner in vivo. By combining absorption and fluorescence LOT into one unit, it also has the potential to study neurovascular coupling since hemodynamic responses and neural activities can be imaged by absorption and fluorescence LOT, respectively. Being able to achieve depth-resolved imaging of both absorption and fluorescence contrast, LOT/MFMT is also a promising nondestructive imaging tool in oncological applications and the tissue engineering area.
We thank Drs. Hyounguk Jang, Jianting Wang, Dmitry Lenkov, and Anna Volnova for their helpful discussions. This work is supported by NIH grants NS039050 (RSE), NS084818 (RSE), R21EB012215 (YC) and R01EB014946 (YC), NSF grant CBET-1254743 (YC), and UMB-UMCP Seed Grant (RSE and YC). The effort in completing this review paper does not involve the use of animal and/or human subjects. The results presented are cited from the published literature noted in the references. The authors declare no conflict of interest.
Qinggong Tang received his BS degree in optoelectronics from Huazhong University of Science and Technology, Wuhan, China, in 2012. He is currently a PhD candidate in the Fischell Department of Bioengineering at the University of Maryland, College Park. His interests include biophotonics and imaging, optical coherence tomography, multiphoton microscopy, fluorescence laminar optical tomography and their applications in neuroscience, and cancer research.
Jonathan Lin is an undergraduate student working toward his BS degree in bioengineering at the University of Maryland, College Park, and will graduate in the spring of 2017. He currently works as a research assistant at the Biophotonics Lab there and is the CEO and cofounder of Reflective Information, Inc., a startup company that develops smart devices with advanced image processing software.
Vassiliy Tsytsarev has a PhD in neuroscience from Saint-Petersburg State University in Russia. Soon after graduation, he moved to Japan and started work in the Brain Science Institute of RIKEN and Kyoto University. After 7 years in Japan, he moved to the United States and is now working at the University of Maryland. Functional brain mapping, neural circuits, and different types of brain optical imaging are his main scientific interests, as well as professional background.
Reha S. Erzurumlu is a professor of anatomy and neurobiology at the University of Maryland, School of Medicine. He received his PhD from the University of California Irvine and did postdoctoral work at Brown University and M.I.T. He is an established investigator with expertise in development, plasticity, and organization of sensory systems. His research uses the rodent trigeminal system as a model and employs numerous approaches ranging from molecular biology, tissue culture, immunohistochemistry, morphology, electrophysiology, imaging, and behavior.
Yi Liu received his BS degree in optoelectronics from Huazhong University of Science and Technology, Wuhan, China, in 2015. Now, he is a PhD student at the Fischell Department of Bioengineering at the University of Maryland, College Park. His scientific interests include bioimaging, optical diagnostic devices, fluorescence laminar optical tomography, and redox fluorescence imaging.
Yu Chen is currently an associate professor and associate chair of the Fischell Department of Bioengineering, University of Maryland, College Park. His research interests include optical coherence tomography, laminar optical tomography, multiphoton microscopy, development of quantitative optical sensing and imaging devices for clinical translation, and preclinical and clinical applications of optical technologies in imaging brain function, renal physiology, cancer therapy, and tissue engineering.