1. INTRODUCTION

Optical imaging has historically played a fundamental role in biological discovery being the mainstay of microscopic observations [1]. While various fluorescence microscopy and diffusion tomography techniques are capable of in vivo deep-tissue imaging [2,3], fundamental physical barriers imposed by optical diffraction and intense photon scattering by biological tissues limit the achievable resolution and depth [4]. Breaking through the optical diffraction barrier was achieved with superresolution approaches such as stochastic optical reconstruction microscopy and photoactivated localization microscopy. Those are based on localizing randomly activated molecules or selective deactivation of fluorophores, allowing the visualization of subcellular structures in previously unattainable detail [5–7]. Three-dimensional imaging was further enabled by exploiting depth dependence of the point spread function (PSF), e.g., formed with a cylindrical lens [8] or biplane detection [9]. Yet, another fundamental resolution and penetration barrier stems from light scattering by tiny inhomogeneities in the refractive index of tissues. The so-called light diffusion barrier largely determines the maximum depth at which “ballistic” photons can be exploited for high-resolution imaging [10]. Nonlinear optical microscopy exploits longer wavelengths of excitation less affected by scattering [11,12], while optical coherence tomography further capitalizes on coherence time gating to reject scattered photons, but the achievable imaging depth in living tissues remains around 1–2 mm [13]. On the other hand, methods based on adaptive optics or wavefront shaping have been suggested to break through this depth barrier [14,15], although in vivo applicability at depths beyond ${\sim}{{1}}\;{\rm{mm}}$ remains challenging.

The recent introduction of efficient short-wave infrared (SWIR) cameras based on InGaAs sensors opened new opportunities for deep-tissue imaging [16,17]. The use of longer excitation/emission wavelengths in the second near-infrared window (NIR-II, ${\sim}{{1000 {-} 1700}}\;{\rm{nm}}$) has extended high-resolution optical imaging toward ${\sim}{{2}}\;{\rm{mm}}$ depths in living tissues by capitalizing on diminished light scattering and autofluorescence at these wavelengths. This previously unachievable performance fostered the development of fluorescent contrast agents with enhanced emission in the NIR-II range [18–24].

 

Fig. 1. Concept of diffuse optical localization imaging (DOLI) and characterization of microdroplets. (a) Layout of DOLI setup. A monochromatic laser beam illuminates fluorescent targets hidden behind the scattering media with backscattered fluorescence light detected by a SWIR camera. (b) WF image of microdroplets captured with a commercial bright-field microscope. (c) Histogram of microdroplet diameter distribution. (d) Localization and image formation workflow. (e) Experimental arrangement for measuring dependence of the PSF on the target depth in a scattering medium. (f) WF image of the microfluidic chip captured with the SWIR camera. (g) The recorded fluorescence spot size (FWHM of the line profiles) as a function of the target depth; both raw data and curve fitting are shown.

Download Full Size | PDF

Deep-tissue imaging with optical contrast can alternatively be accomplished with hybrid approaches combining light and sound. In particular, ultrasound waves experience little scattering in soft biological tissues as compared to light, thus several acousto-optic methods were proposed employing focused ultrasound to modulate coherent light and create a frequency-shift light source inside a turbid sample. Then, detection of scattered wavefront serves to refocus light into the acoustic focus through time-reversal optical phase conjugation [25,26]. However, these methods are subject to millisecond-scale speckle decorrelation times in living tissues, hindering the potential in vivo applications. Another concept termed fluorescence and ultrasound-modulated light correlation was proposed based on a high correlation between ultrasound-tagged light and fluorescence fluctuations localized within the same voxel inside an opaque sample [27]. In addition, optoacoustic (OA) imaging, where ultrasound is generated via absorption of light pulses, has become an established tool in biomedical research [28–30]. The penetration and spatial resolution of optical-resolution OA microscopy approaches employing focused excitation beams [31,32] are similarly limited by the light diffusion barriers. OA imaging at centimeter-scale depths is possible when employing OA imaging at near-infrared wavelengths in the so-called acoustic-resolution regime with unfocused light excitation [33,34]. In the latter case the spatial resolution scales by a factor of approximately 1/200 of the imaging depth [28]. Breaking through the acoustic diffraction barrier has recently been enabled with localization-based techniques, such as the ultrasound localization microscopy [35] and localization optoacoustic tomography [29]. Note that the OA methods generally differ from the fluorescence-based techniques in that the image contrast is mainly associated with hemoglobin absorption, which may compromise sensitive detection of extrinsic labels in the presence of strong background absorption by blood [36].

Herein, we introduce diffuse optical localization imaging (DOLI) to overcome the barriers imposed by photon scattering. The method capitalizes on localization imaging principles to accurately track flowing microdroplets encapsulating lead sulfide (PbS)-based quantum dots in a sequence of epi-fluorescence images acquired with a SWIR camera in the NIR-II spectral window, thus enabling high-resolution fluorescence imaging in the diffuse regime of light.

2. MATERIALS AND METHODS

A. Epi-Fluorescence Imaging System

DOLI is implemented with a typical epi-fluorescence (widefield, WF) system configuration. The layout of the experimental setup is depicted in Fig. 1(a). A continuous-wave laser diode was employed for excitation of the sample. The laser beam was diverged to adjust the average power intensity to ${{850}}\;{\rm{mW/c}}{{\rm{m}}^2}$, which exceeds by a factor of ${\sim}{{2}}$ the permissible levels of human skin exposure at the 855 nm wavelength of the laser [37]. The emitted fluorescence from the sample was filtered with a long-pass filter (FEL1100, Thorlabs, USA) and magnified (${\sim}{{0}.\rm{35}\! -\!{\rm 0}.{50}}$ ratio) with a NIR-II coated camera lens (LM50HCSW, 50 mm effective focal length, Kowa, Japan). The fluorescence images were recorded with an InGaAs-based SWIR camera (WiDy SenS 640V-ST, NiT, France) operated in the linear mode under ${-}{{15}}^\circ$ air cooling. The pixel pitch of the SWIR camera is 15 µm at full resolution of ${{640}} \times {{512}}$ pixels without the additional binning function.

B. Microdroplet Synthesis and Characterization

Synthesis of microdroplets was performed following a standard emulsification procedure. Toluene droplets were used in the phantom experiments. These were prepared by adding 50 µL 10 mg/ml PbS core-type quantum dots (PbS QDs) in toluene (747076, 1400 nm emission, Sigma-Aldrich, Germany) to 1 ml DI water containing 3% v/v TWEEN20 surfactant (P1379, Sigma Aldrich, Germany). The mixture was subsequently vortexed for 30 s and filtered with a Falcon 40 µm cell strainer to result in a relatively uniform microdroplet suspension. The microdroplet size was characterized with a bright-field microscope (Axioskop, Carl Zeiss, Germany). WF images of the microdroplets distributed on a microscopic slide were captured [Fig. 1(b)]. The diameter of each isolated microdroplet was analyzed with ImageJ build-in functions [38]. The histogram of droplet size was shown in Fig. 1(c) with a mean value and standard deviation of 8.41 µm and 4.50 µm, respectively. In vivo experiments were performed with dichloromethane (DCM) droplets, which were prepared with core/shell lead sulfide/cadmium sulfide quantum dots (PbS/CdS QDs) in DCM (NBDY-0038, 1600 nm emission, Nirmidas Biotech, USA) following the same emulsification procedure described above.

C. Phantom Imaging

A microfluidic chip (10000212, Microfluidic ChipShop, Germany) with a 100 µm diameter single channel was used in phantom experiments. These were prepared with 1.2% v/v Intralipid (I141, Sigma Aldrich, Germany) suspended in agar, mimicking the average optical scattering in biological tissues for the relevant wavelength range [39]. Following a linear relationship between the reduced scattering coefficient and concentration of Intralipid, 1.2% v/v Intralipid concentration corresponds to $\mu _s^\prime = {17.28}\;{\rm{c}}{{\rm{m}}^{- 1}}$ at 633 nm [40]. The phantoms consisted of Intralipid/agar layers with thicknesses of 0 mm, 1 mm, 2.5 mm, and 4.0 mm, placed on top of the microfluidic chip to mimic light propagation through the corresponding depth in tissues. The microdroplet suspension composed of PbS QDs in toluene was injected through a single channel on the chip. The flow velocity was controlled with a syringe pump to reduce motion blurring. A sequence of epi-fluorescence (WF) images was recorded for each phantom by adjusting exposure time of the SWIR camera. Specifically, 10 ms exposure time was used for the phantoms with thicknesses up to 2.5 mm, while 30 ms was used for the fourth phantom (4 mm thickness). Each sequence corresponded to a total acquisition time of 5 min.

For the resolution characterization experiments, images acquired from a single microfluidic channel were superimposed to generate a “virtual” crossing channel. This was achieved by rotating each subframe recorded from the single channel 10° and ${-}{{10}}^\circ$ around its center and summing up the corresponding sequences with a time delay of 100 frames so the microdroplet positions remain uncorrelated. The virtual crossing approach was selected because flowing particles are commonly arrested at the intersection part when using a microfluidic chip with two physically crossing channels. Since the camera is merely used to track individual particles and has a linear response to fluorescence signals, the two approaches are expected to render equivalent results.

D. In Vivo Imaging

Three athymic nude-Fox1nu mice (Envigo RMS B.V., Netherlands) were used for in vivo measurements. The mice were anesthetized with isoflurane (3% v/v for induction and 1.5% v/v for maintenance) in a mixture of medical air and oxygen with flow rates of 0.8 L/min and 0.2 L/min, respectively. For transcranial microangiography of the mouse brain, the scalp of a nude mouse (10-week-old) was removed after subcutaneous injection of analgesics (Buprenorphine, 0.1 mg/kg), followed by intravenous injection of 100 µL microdroplets suspension made of PbS/CdS QDs in DCM. A time-lapse image stack was recorded for 5 min during and after injection to form the DOLI image. The exposure time of the SWIR camera was set to 20 ms with effective frame rate of 41 Hz. Overall, 12,300 successive frames were recorded to reconstruct the DOLI image. An epi-fluorescence image was subsequently acquired with the same camera exposure time following injection of 25 µL aqueous-soluble PbS/CdS QDs at 5 mg/ml in phosphate-buffered saline (NBDY-0018, 1600 nm emission peak, Nirmidas Biotech, USA) for verification and comparison with DOLI. Fully noninvasive mouse brain vascular imaging was also performed. For this, the DOLI image of the second mouse (5-week-old) with scalp intact was acquired post microdroplet injection following the same procedures mentioned above. The exposure time of the SWIR camera was increased to 50 ms with an effective frame rate of 18 Hz for improving the signal-to-noise ratio (SNR). A time-lapse image stack consisting of 10,800 frames was recorded for 10 min post injection. Another DOLI image was collected for the same mouse after scalp removal. This served as a reference to distinguish vessels from the brain and the scalp. As a comparison, a WF image of the mouse brain through the intact scalp was acquired for another mouse following injection of 25 µL 5 mg/ml aqueous-soluble PbS/CdS QDs. After image acquisition, all mice were euthanized while still under anaesthesia. All animal experiments were performed in accordance with the Swiss Federal Act on Animal Protection and approved by the Cantonal Veterinary Office Zurich.

E. Localization-Based Image Reconstruction and Depth Mapping

The DOLI image formation workflow is depicted in Fig. 1(d). For each recorded time-lapse dataset, singular value decomposition (SVD) filtering [41,42] was first applied to extract the signal fluctuations corresponding to flowing microdroplets from the static background. After SVD of the space–time matrix, a threshold separating the singular vectors corresponding to the background from those corresponding to flow was applied. The maximum intensity projection of the WF image stack was used as a reference to adjust this threshold. Subsequently, the center of single microdroplets in each filtered subframe was located by searching for local maxima with an adaptive threshold method. In addition, the spot size of each recognizable target was determined as the full width at half-maximum (FWHM) after 2D Gaussian fitting. The trajectory of each detected microdroplet was traced with a Simpletracker algorithm [43]. The maximum linking distance, which determines the searched region of the flowing microdroplets in consecutive subframes, was adjusted according to the actual flow velocity and frame rate. In addition, a threshold was set to the length of trajectories to eliminate potential false trajectories or stacked microdroplets. The final DOLI intensity image was formed via superimposition of centroid trajectories. Besides the intensity map, color-coded spot size images were simultaneously formed by superimposing the estimated spot size to the localized points in each subframe. This further provided a rough depth estimation for each detected microdroplet. Similarly, with the given frame rate and relative displacement of microdroplets in consecutive frames, the velocity and direction of each pixel were calculated to form a color-encoded flow velocity and direction map.

 

Fig. 2. Resolution enhancement achieved with DOLI in diffuse media with respect to the conventional WF images. Panels (a)–(d) correspond to different thicknesses (0–4 mm) of the Intralipid/agar phantoms placed on top of the microfluidic chip channel as indicated. The image profiles for the WF and the DOLI images along the indicated green lines are shown in the right column. Bilinear interpolation was applied here to alleviate pixelation effects. (e) DOLI images reconstructed for different thicknesses of the Intralipid layer. Color coding corresponds to the depth of flowing QDs that was estimated based on the reconstructed spot size [Fig. 1(g)]. (f) Mean and standard deviation (SD) values of fitted depth across the field of view plotted versus actual thickness of the Intralipid layer.

Download Full Size | PDF

3. RESULTS

A. Imaging Through Scattering Phantoms with DOLI

To calibrate the spot size against the corresponding depth information, a 50 µm diameter single-channel microfluidic chip (10000210, Microfluidic ChipShop, Germany) filled with aqueous-soluble PbS/CdS QDs was placed under a customized ladder-shape Intralipid–agar layer (1.2% v/v). The Intralipid/agar phantom was used to vary the target depth range between 0 to 3.5 mm [Fig. 1(e)]. WF images of the microfluidic chip were captured by the SWIR camera with 100 to 220 ms exposure time depending on the SNR of the images [Fig. 1(f)]. An averaged image of 1000 frames acquired at 4 Hz frame rate was used to achieve an optimal SNR. Line profiles along the vertical direction to the depth gradient were extracted pixel by pixel, followed by Gaussian fitting to obtain the FWHM value which was considered as the spot size at the given depth. During this process, we assumed that the influence of fluorescence originating from neighboring pixels is negligible. In this way, a relationship between the practical spot size and scattering depth was established, followed by curve fitting to a fourth polynomial [Fig. 1(g)]. Note that the curve shape is generally strongly dependent on scattering properties of the medium (Fig. S1, Supplement 1).

As an initial proof-of-principle demonstration of spatial resolution enhancement with DOLI, flowing microdroplets were imaged as previously described. Figures 2(a)–2(d) show a comparison of the equivalent WF images obtained by superimposing all frames from the acquired sequence and the DOLI images rendered by superimposing the localized position of the droplets. Note that the images represent a “virtual” cross (see Section 2). The intensity profiles along the green line indicated in each group are also shown [Figs. 2(a)–2(d), right]. Both normal WF and DOLI images feature similar intensity profiles when no scattering layer is present [Fig. 2(a), left], although a better SNR is achieved with DOLI [Fig. 2(a), right]. Increasing light scattering results in smearing of the two channels in the WF images, while the corresponding DOLI images remain almost unaffected [Figs. 2(b) and 2(c)]. The benefits of DOLI become obvious beyond a certain depth. For example, no clearly defined structures can be distinguished with WF over the entire field of view (FOV) under a 4.0 mm thick Intralipid layer [Fig. 2(d)], while the 198.3 µm interval between the crossing channels can still be unambiguously resolved with DOLI. The depth information for different thicknesses of the Intralipid layer [Fig. 2(e)] was then derived using the estimated relation between the reconstructed spot size and depth [Fig. 1(g)], correlating well to the actual thickness of the Intralipid layer [Fig. 2(f)].

 

Fig. 3. Transcranial cerebrovascular mapping in mice with DOLI. (a) WF image of fluorescent dye perfusion through the murine cerebrovascular system after scalp removal. (b) Corresponding DOLI image acquired for the same mouse following intravenous injection of the microdroplet suspension. (c),(d) Zoom-in views of ROIs indicated in (a) and (b). SSS, superior sagittal sinus; ACA, anterior cerebral artery; MCA, middle cerebral artery; TS, transverse sinus.

Download Full Size | PDF

Under ideal (noise free) conditions, the spatial resolution of DOLI is chiefly determined by pixel size and frame rate of the camera used to localize positions of individual flowing particles, in principle being independent from imaging depth within the scattering medium. In practice, the localization accuracy is affected by the SNR degradation with increased scattering, thus the effective DOLI resolution accordingly decreases with depth, as is evident from Fig. 2(d).

B. In Vivo Microangiography of the Mouse Brain

The DOLI performance is further showcased by high-resolution transcranial cerebrovascular imaging in mice. For this, the WF image of the mouse brain recorded following intravenous injection of aqueous PbS/CdS QDs was taken as a reference after scalp removal [Fig. 3(a)]. The central vascular branch was clearly depicted in this image owing to the suppressed scattering in the NIR-II window. The DOLI image for the same mouse was rendered from the sequence of images acquired after intravenous injection of the microdroplet suspension [Fig. 3(b)]. Zoom-in images of a representative region of interest (ROI) are also shown to better illustrate the significant spatial resolution and contrast enhancement with DOLI [Figs. 3(c) and 3(d)]. The microvascular features are clearly more visible in the DOLI image.

Generally, extensive light scattering by the scalp is known to significantly challenge the performance of optical imaging techniques. Previous studies demonstrating noninvasive cerebrovascular imaging with high-quantum-yield NIR-II contrast agents evinced the hard trade-off between spatial resolution and depth [19,22,23]. A typical WF image of the mouse brain through the intact scalp captured after intravenous injection of aqueous PbS/CdS QDs is displayed in Fig. 4(a). Blurring at the edge of the image is ascribed to the increase in effective tissue thickness along the vertical direction. By randomly distributing the neighboring fluorescence emitters between separate time-lapse frames, DOLI renders background-free high-resolution microvascular map over extended FOV in a totally noninvasive manner [Fig. 4(b)]. In this image, the vascular network in the scalp is superimposed onto deeper situated cerebral vasculature. Naturally, the scalp vasculature is not present in the images acquired with the scalp removed, making the brain vessels easily discernible [Fig. 4(c)]. We subsequently superimposed DOLI images acquired with and without the scalp in contrasting colors [Fig. 4(d)]. Typical cerebrovascular structures such as the superior sagittal sinus and middle cerebral arteries are clearly visible in both images. Note that the smallest resolvable feature is mainly determined by the size of vessels where microdroplets of the given size are allowed to circulate as well as any optical heterogeneities that may lead to distortion of the PSF and thus affect accuracy of the centroid/maxima-based localization.

 

Fig. 4. Noninvasive cerebrovascular mapping in mice with DOLI. (a) WF image of the mouse head through the intact scalp after fluorescent dye perfusion. (b) Corresponding DOLI image formed from the image stacks acquired after the microdroplet injection. (c) DOLI image of approximately the same ROI acquired after scalp removal. (d) Combined microvascular map of the brain and scalp by superimposing DOLI images with and without scalp. ICV, inferior cerebral vein; SSS, superior sagittal sinus; MCA, middle cerebral artery; TS, transverse sinus. (e) Representative time-lapse images of microdroplets from three ROIs indicated with the solid orange squares in (b). (f), (g) Color-coded DOLI depth maps recorded with and without scalp, respectively. Depth estimation was based on the spot size to depth calibration curve shown in Fig. 1(g). (h) Zoom-in views of ROIs indicated with the dashed white squares in (f) and (g). (i) Depth statistics (${\rm{mean}}\;{{\pm}}\;{\rm{SD}}$) in selected ROIs, as indicated with white solid squares in (f) and (g).

Download Full Size | PDF

As previously shown in the phantom experiments, the size of the image spots corresponding to circulating droplets is strongly influenced by the scattering properties of tissue and the depth of field (DOF) of the camera lens, which can be exploited for discriminating structures located at different depths. Representative images of microdroplets circulating in vessels from different depths are shown in Fig. 4(e). For the given DOF of the camera used in the experiment (${\sim}{{4}}\;{\rm{mm}}$), broadening of the spot size is mainly related to the depth within a scattering medium. Note that magnification was kept constant for all the measurements. Subsequently, the derived approximate exponential broadening formula [Fig. 1(g)] was applied to in vivo data to map the depth information in the reconstructed DOLI image with the scalp intact [Fig. 4(f)]. The yellow- and red-colored vessels in Fig. 4(f) correspond to cerebral vasculature which is located deeper than the scalp vasculature shown in blue–green colors. A good correspondence exists between the color-coded DOLI depth map rendered with the scalp intact [Fig. 4(f)] and the superimposed DOLI image [Fig. 4(d)], further confirming the validity of spot size quantification as a depth estimation method. For comparison, the DOLI depth map obtained after scalp removal is shown in Fig. 4(g) with the zoom-in views of selected ROIs shown in Fig. 4(h). As expected, the estimated depth of brain vessels is higher when the scalp is present [Fig. 4(i)]. The average difference in the depth of brain vessels estimated from DOLI images acquired with and without scalp is ${\sim}{{700}}\;{{\unicode{x00B5}{\rm m}}}$, in congruence with the reported scalp thickness values [44]. Overall, DOLI images are shown to clearly improve the SNR and spatial resolution over conventional in vivo WF brain images while additionally providing depth information.

 

Fig. 5. Velocity and flow direction maps estimated by tracking flowing microdroplets in consecutive frames with DOLI. (a) Reconstructed velocity maps with color-encoded velocity in the range of 0–12 mm/s. (b) Reconstructed direction map with color-encoded angle indicated by the color-wheel.

Download Full Size | PDF

In addition to the resolution improvement, DOLI can be exploited to retrieve additional functional information on blood flow velocity and direction by tracking the microdroplet trajectories. As expected, major brain vessels exhibit higher blood velocities compared to smaller vessels in the scalp [Fig. 5(a)]. In addition, a color-encoded flow direction map was reconstructed [Fig. 5(b)], where a symmetrical flow in the two hemispheres is observed. Note that the relatively slow frame rates and large microdroplet size in the current implementation are not ideal for accurate estimation of the blood flow velocity and direction using the localization approach [45].

4. DISCUSSION AND CONCLUSIONS

The presented results demonstrate that DOLI greatly enhances the resolution and penetration capabilities of conventional WF fluorescence imaging. The localization imaging concept was shown to successfully preserve spatial resolution of fluorescence imaging across 4 mm thick turbid tissue phantoms when employing microdroplets featuring fluorescence emission in the NIR-II window. DOLI further demonstrated noninvasive, high-resolution cerebrovascular imaging in living mice, where visibility of cerebrovascular structures with WF is significantly hampered by the strong light scattering within the murine scalp and skull.

Several issues need to be considered for fully exploiting DOLI’s potential as a new biomedical imaging tool. First, the maximum size of fluorescent particles used for localization should be limited to about 5 µm so they can freely propagate throughout microvascular networks without causing capillary arrest. This implies the use of fluorescent emitters with superb brightness so they can be individually detected deep in scattering and absorbing tissues. A growing number of contrast agents featuring fluorescence emission in the NIR-II are becoming available [20–24,46–48], thus expected to facilitate imaging with DOLI. Another key aspect to be optimized is the particle concentration. Theoretically, the localization accuracy is determined by the SNR of the images rather than by widening of the PSF caused by light scattering and the DOF of the imaging system. However, particles need to be sparsely distributed and provide sufficient SNR to be individually localized both in space and time. Thereby, the particle concentration as well as the camera sensitivity and dynamic range for detecting particles from different depths impose practical limitations on the achievable depth. DOLI can also be used in the visible range [45] and first near-infrared window, where more sensitive and faster cameras as well as brighter contrast agents are currently available, albeit the achievable depth is generally expected to be inferior in this spectral range. In this work, the depth estimation relied on an estimated formula linking the spot size to the target depth in the scattering medium, which was experimentally measured with a relatively simple tissue-mimicking phantom. To enhance the accuracy of depth estimation, a more delicate tissue modeling approach can instead be undertaken to more accurately account for heterogeneous scattering and absorption properties of living tissues.

Additional technical advances may further boost the performance of DOLI. For example, three-dimensional localization microscopy has been realized with sophisticated optical designs exhibiting depth-dependent PSF, e.g.,  based on cylindrical lenses or double-helix phase masks [49,50]. A similar approach can be applied to improve the depth estimation of particles in the current configuration. A combination of images from two (or several) cameras with different orientations with respect to the sample may alternatively be used for three-dimensional imaging. In addition, accurate modeling of light diffusion through tissues may facilitate enhanced localization accuracy in three dimensions [51]. Other diffusion-based optical imaging techniques, such as fluorescent molecular tomography [52], could potentially benefit from the newly introduced approach by boosting the resolution and accuracy of targeted fluorescent agent reconstructions for deep-tissue molecular imaging applications.

In conclusion, we expect that DOLI emerges as a powerful approach for fluorescence imaging of living organisms at previously inaccessible depth and resolution regimes, thus massively enhancing the in vivo applicability of fluorescence microscopy and tomography techniques.