1. Introduction

The image grabbing speed is of fundamental importance in practical biology applications, such as in-vivo microenvironment analysis, living organism research, and neuron science [1–3]. In particular, a rapid volumetric imaging capability is essential to realize dynamic observations [4–6]. Many rapid point-scanning multiphoton excitation microscopy (MPEM) techniques have been proposed for achieving a dynamic volumetric imaging capability using high-speed elements such as acousto-optic scanners [7], high-speed scanning polygonal mirrors [8], and resonance mirrors integrated with electrically-tunable lenses or tunable acoustic lenses [9–11]. The frame rate of MPEM systems can be further improved through the use of widefield excitation microscopy to illuminate the entire two-dimensional (2D) area of interest. Several studies have employed light-sheet microscopy with plane-projection techniques to capture whole brain functional images of zebrafish [12,13]. Similarly, the authors in [14] and [15] obtained high-resolution dynamic images using multiple-beam lattice light-sheet microscopy and Bessel-beam excitation with a pair of axicons, respectively. However, the light illumination and emission collection directions in light-sheet one-photon microscopy are mutually orthogonal, which may cause inconvenient in implementation or alignment with in vivo applications. Consequently, the feasibility of performing widefield excitation through temporal focusing (TF) techniques with MPEM has attracted growing attention [16]. Compared to the methods in [12–15], TF-MPEM can be implemented using normally forward illumination and normally backward collection. As a result, being more intuitive and flexible. Conventionally, TF is realized by using gratings [17–21], spatial-light-modulators (SLMs) [22], and digital micromirror devices (DMDs) [23,24] to disperse the femtosecond laser into its wavelength components, while these wavelength components overlapping in the spatial again, only the plane that they are temporal focused has the enough photon density to excite the florescence. Besides the plane-projection capability of TF, which provides a widefield excitation capability, the TF plane can be remotely scanned using a dual-prism grating or group delay dispersion device to add a phase delay in the axial scanning direction, thereby enabling high-speed volumetric imaging [23,25,26].

To further improve the volumetric imaging rate of the systems, it is desirable to eliminate the mechanical scanning process and obtain the volumetric image directly using some form of one-snapshot process. One of the most common techniques for acquiring one-snapshot records of overall 3D information is that of light-field microscopy (LFM) [27,28]. The main concept of LFM is to utilize four dimensional coordinates to describe the 3D light propagation behavior recorded in a 2D images. In LFM systems, the image is typically captured using a microlens array (MLA) or attenuating mask placed at an appropriate distance from an image sensor [29,30]. For light-field (LF) cameras or microscopes with a low numerical aperture (NA) [31–33], the 3D scene information can be rapidly analyzed and applied to construct 3D images using paraxial and geometrical optics [34]. However, in high NA systems, wave optics theory must be employed to process the LF image data at the microscopic level [35]. Moreover, the Debye model must also be used to evaluate the point spread function (PSF) in the incoherent image system [35–37].

Several high NA LFM systems with lateral and axial resolutions on the order of just several microns have been successfully developed [38,39]. However, the ability to suppress the background noise caused by biotissue scattering, and enhance the imaging quality of the captured volume, remains an important concern. Recently, several studies have combined single photon light-sheet illumination techniques with LFM (i.e., the volume of interest (VOI)) [40–42] to reduce background noise and increase image contrast by selecting the illumination volume [40]. In the light-sheet illumination, the illumination path is perpendicular to the detection path. The present study thus proposes a temporal focusing multiphoton illumination (TFMI) system with selective volume illumination (SVI) capability to illuminate only the VOI and to adjust the 3D excitation confinement effect as required to realize deep tissue penetration. Compared to the single photon light-sheet illumination, the present TFMI approach provides an epifluorescence setup (i.e., a normal-illumination and normal-detection configuration), which is thus more intuitive and easily-implemented than other methods for most bioimaging applications. Furthermore, the TFMI has the advantages of multiphoton excitation, i.e., deeper penetration depth and less photobleaching effect [43].

The TFMI-SVI method is embedded in an LFM system by placing two lens pairs with different magnifications in front of the objective in order to adjust the volume illumination range. Moreover, a LF phase-space deconvolution algorithm for high NA systems [28,39] is employed to reconstruct the LFM images directly into 3D volumetric images without the need for mechanical scanning. It is shown that the TFMI-LFM system achieves lateral and axial resolutions near the native image plane of around 1.1 µm and 1.2 µm, respectively, when applied to the imaging of 100-nm fluorescent microbeads. The biotissue imaging capability of the proposed system is demonstrated by comparing the images acquired of the single lobe of the drosophila mushroom body (MB) with genetic fluorescent marker (GFP) with those captured by a TF-MPEM system with axial scanning. It is shown that similar mesh features are obtained in both cases. Accordingly, the proposed TFMI-LFM system appears to provide a promising approach for the rapid volumetric imaging of biotissues with high spatial resolution. Compared to the conventional light field microscopy, our approach has less background noise and better contrast. Moreover, the present study, which based on the multiphoton epifluorescence setup, provides deeper penetration depth and lower photobleaching, and is more intuitive and easily-implemented for biotissue imaging.

2. System and methods

2.1 TFMI-LFM and TF-MPEM systems

Figure 1 presents a schematic illustration of the overall TFMI-LFM and TF-MPEM setup. As shown, the illumination light is provided by a high-power femtosecond amplifier (Pharos PH1-10, Light Conversion) with a 1030-nm excitation wavelength. Moreover, lenses L_3L and L_4L (or L_3T and L_4T) comprise a relay set, which together with lens L₅ and the objective (Olympus UPLSAPO60XW 60x/1.2), form a 4-f system for TF illumination in an upright optical microscope (Axio Imager.A2m, Carl Zeiss). The MPEM signal emitted from the sample is filtered by a dichroic mirror and short-pass (SP) filter to remove the reflected excitation laser light and is then imaged to either a high-sensitivity EMCCD camera (iXonEM + 897 EMCCD, Andor) through an imaging lens for TF-MPEM imaging or a high-sensitivity EMCCD camera (iXonEM + 885 EMCCD, Andor) through the imaging lens and an LF lens set for TFMI-LFM imaging. The EMCCD 897 and 885 cameras have 512 × 512 active pixels with a 14-µm pixel size and 1,002 × 1,004 active pixels with an 8-µm pixel size, respectively. Both cameras can be thermoelectrically cooled down to −80°C in air and provide a 14-bit digitization resolution and maximum 35-MHz readout rate. For the TF-MPEM, 3D images are acquired by controlling a motorized stage (H101A ProScan, Prior Scientific) with a 3-axis encoder and a rapid piezo focusing stage (NanoScanZ 200, Prior Scientific) with a maximum travel range of 200 µm. For both systems, all of the peripheral instrument communications and control functions are handled through a high-speed data acquisition card (PCIe-7842R, National Instruments) and Virtex-5 LX50 FPGA under the control of a self-written LabVIEW program.

 

Fig. 1. Schematic representation of TFMI-LFM and TF-MPEM system setup. The left-hand side in the dashed rectangle shows the light field imaging system, while the right-hand side shows the TFMI system with a volume selective lens pair (L_3L & L_4L) and original lens pair (L_3T & L_4T).

Download Full Size | PDF

According to TF theory [19], the axial excitation confinement (AEC) of the illuminating light in TF-MPEM systems is governed by the width of the laser beam on the back focal plane of the objective lens. Thus, in the present setup, a simple two-lens-pair relay system operated under the control of a flip mirror is used to switch between the original TF-MPEM system and the proposed TFMI-LFM system, respectively. As shown in Fig. 1, the equivalent focal length of the L_3L-L_4L lens pair for the TFMI-LFM system is shorter than that of the L_3T-L_4T lens pair for the TF-MPEM system. As a result, the light passed through the TFMI-LFM lens pair has a narrower beam width on the back focal plane of the objective, and thus produces a wider AEC. (See Sec. 2.2). In the TFMI-LFM lens setup, a further pair of relay lenses are positioned between the microscope port and the MLA (MLA-S100-F21, RPC Photonics) in order to adjust the image plane of the microscope such that it is imaged exactly on the MLA. Furthermore, two camera lenses (AF-S NIKKOR 50MM F/1.8G and AF-S NIKKOR 85MM F/1.8G) are coupled nose-to-nose and used to relay the image from the back focal plane of the MLA to the sensor surface of the EMCCD 885 camera. Notably, the use of camera lenses rather than common optical lenses reduces the aberration curve of field and hence improves the imaging quality of the TFMI-LFM system.

2.2 LFM image reconstruction

Fluorescent images are incoherent with an independent phase among the object points. Consequently, the LF system can be regarded as an incoherent image formation system. In incoherent image formation, the PSF can be utilized to describe the optical system, and the intensity is thus given by the modulus square of the amplitude [35]. The overall PSF on the MLA under high NA conditions can be evaluated using scalar Debye theory [35–37,44]. Furthermore, the behavior of the light field propagating from the MLA to the image sensor can be modeled using the Fresnel approximation over the focal length of the MLA [37,44]. In general, the PSF represents the model which projects fluorescence light from the volume of the object to camera pixels. Due to the nonuniform effect of the MLA, point sources are produced at different positions, and these point sources then induce different PSFs. Therefore, the PSF matrix must include five dimensional coordinates, namely two coordinates for the pixels located on the image plane and three coordinates for the voxels constituting the object volume space. Herein, the angular spectrum method is employed to model the PSF matrix of the LFM system from a point source at position (x, y, Δz) to the MLA plane. In addition, the back-focal aperture size of the objective lens is modeled as:

With the evaluation of the PSF matrix, the volume reconstruction problem can be regarded as a form of inverse problem [44]. In the present study, volume reconstruction is performed using the phase-space deconvolution method [45], which is developed from the traditional 3D deconvolution approach. The 3D deconvolution method utilizes the well-known Richardson-Lucy (RL) algorithm [46] and solves the inverse reconstruction problem iteratively. By contrast, the phase-space deconvolution method realigns the LFM raw image and PSF matrix intensity, ${|{PS{F_{CCD}}({x,\; y,\; \Delta z} )} |^2}$ according to the CCD pixel position related to each lens of the MLA to form a new realigned image and matrix, and then operates a process similar to the RL algorithm using the new realigned image and matrix. The simulation results demonstrate that the reconstructed image obtained via phase-space deconvolution is more uniform than that produced by the original 3D deconvolution method and has a superior resolution and contrast in both the lateral and axial spaces. Recently, several studies have applied neural networks to solve the inverse problem [41,42,47,48] based on a large number of training samples, and have achieved well-refocused images

3. Experimental results and discussions

3.1 TFMI with selective volume illumination

Figure 2(a) shows the simulated variation of the AEC with the laser beam width at the back focal plane of the objective in the system shown in Fig. 1. The orange curve in Fig. 2(b) shows that the original TF-MPEM system has an AEC full width at half maximum (FWHM) of 2.9 µm. Note that the normalized intensity profiles in Fig. 2(b) are obtained by measuring the axial intensity profile of a thin film doped with rhodamine 6G (R6G) dye (< 200-nm thick), and the FWHM values of the fitted intensity profiles [19] are estimated without back-focal aperture restriction (i.e., a laser beam width of more than 8-mm fully fills the back-focal aperture of the objective in one direction.) With full-filling of the back focal plane of the objective, the TF-MPEM experimental result for the AEC is close to the simulation result in Fig. 2(a). As described in Sec. 2.1, a SVI capability is introduced into the LFM imaging process by adjusting the beam width at the back focal plane using a relay system comprising lenses L_3L-L_4L and L_3T-L_4T. By selecting lenses L_3T-L_4T, the beam width at the back focal plane of the objective is around three times narrower than that in the original TF-MPEM system. As shown in Fig. 2(b), the FWHM of the resulting normalized intensity profile (21.6 µm) is around seven times larger than that of the intensity profile in the TF-MPEM system (2.9 µm). Thus, the results confirm the SVI capability of the proposed TFMI-LFM system.

 

Fig. 2. (a) Simulation results for variation of AEC FWHM with laser beam width at back focal plane of system setup shown in Fig. 1. (b) Experimental AECs with FWHMs of 2.9 µm and 21.6 µm corresponding to laser beam widths of 8 mm and 2.7 mm, respectively. (c) Intensity profiles of red lines in 2(f) and 2(j). Orange and blue curves correspond to AEC FWHMs of 2.9 µm and 21.6 µm, respectively. (d) LFM raw image for AEC with FWHM of 2.9 µm, and (e)-(g) reconstructed images for refocused planes located 1 µm above, at, and 1 µm below the focal plane, respectively. (h) LFM raw image for AEC with FWHM of 21.6 µm, and (i)-(k) reconstructed images for refocused planes located 1 µm above, at, and 1 µm below the focal plane, respectively.

Download Full Size | PDF

The SVI effect was evaluated by exciting 2-µm florescent beads fixed in agar gel with different volume sizes. Figures 2(d)–2(g) show the LFM raw image and three reconstructed images obtained using an AEC with a FWHM of 2.9 µm. Meanwhile, Figs. 2(h)–2(k) show the corresponding images obtained using an AEC with a FWHM of 21.6 µm. The effect of the SVI is apparent even in the raw image in Fig. 2(h), in which the signal of the defocused beads is clearer than that in Fig. 2(d). The background noise in Fig. 2(h) is more pronounced than that in Fig. 2(d). However, despite this, the wider excitation volume still yields more information about the sample (i.e., the beads located at different depths in the gel can be more clearly observed). Figures 2(e) and 2(i) show the reconstructed images for the refocused plane at a distance of 1 µm above the focal plane. Similarly, Figs. 2(f) and 2(j), and 2(g) and 2(k), show the reconstructed images at the focal plane and at a refocused plane 1 µm below the focal plane, respectively. Figure 2(c) compares the intensity profiles of the red lines in Figs. 2(f) and 2(j). It is seen that the blue curve, which corresponds to the red line in Fig. 2(f), has a better intensity contrast for the first bead than for the second bead. This can mostly likely be attributed to the low FWHM of the excitation signal (2.9 µm), which prevents the full excitation of the bead located at a greater depth in the gel. However, Fig. 2(j) shows that the dim bead is successfully reconstructed when using a larger excitation volume with a FWHM of 21.6 µm. (Note that both intensity signals in Fig. 2(c) are normalized.) Herein, the florescent beads are fixed in clear agar gel, and thus only minimal background noise is induced in both cases However, the reconstruction results in Figs. 2, respectively, nevertheless show subtle differences. In other words, the results confirm that, in biotissue imaging, if not only the VOI is excited, such as the whole sample illumination, scattering effects and florescence signals from depths other than the imaged depth increase the background noise, and thus degrade the reconstructed image contrast. In this experiment, the exposure time of both Figs. 2(d) and 2(h) is only 10 ms for an excitation volume of 200 × 200 × 20 µm³ with the imaging volume rate of 100 Hz under the laser power of 20 mW. While increasing the power to 180 mW, the volume rate can be further promoted to 300 Hz (i.e., $100 \times \sqrt {{{180} / {20}}}$ Hz). However, an over-power illumination could potentially result in the photobleaching effect.

3.2 Spatial resolution of TFMI-LFM

To investigate the spatial resolution of the TFMI-LFM system, 100-nm florescent beads were fixed in agar gel on a cover glass slide and then imaged using an AEC with a FWHM of 21.6 µm. Figures 3(a)–3(c) show the raw TFMI-LFM images obtained when positioning the top surface of the glass slide at the front focal plane of the TFMI-LFM system and at locations 5 µm above and 10 µm above the front focal plane, respectively. Figures 3(d)–3(f) show the corresponding reconstructed images obtained over a depth range of 0 to 10 µm above the front focal plane at intervals of every 1 µm. Figure 3(g) shows the lateral intensity profiles of the three colored squares in Figs. 3(d)–3(f), where these squares contain 100-nm florescent beads located on (blue), 5 µm above (red), and 10 µm above (orange) the front focal plane, respectively. The blue, red, and orange fitted curves have FWHM values (i.e., lateral resolutions) of 1.2 µm, 1.5 µm, and 2.1 µm, respectively. Figure 3(h) shows the intensity profiles of the three squares in the axial direction with 0.5 µm layer distance. The fitted FWHM values (i.e., axial resolution values) are equal to 1.1 µm, 2.4 µm, and 4.5 µm for the beads located on, 5 µm above, and 10 µm above the front focal plane, respectively. The results presented in Figs. 3(g) and 3(h) show that the spatial resolution of the TFMI-LFM system deteriorates with an increasing distance from the front focal plane. Since the low resolution and non-significant change of the intensity in the reconstruction of the beads located on 10 µm above, a little shift along the z axis can be observed in Fig. 3(f). Overall, the system has an axial excitation volume of around 40 µm (AEC with a FWHM of 21.6 µm) with a spatial resolution of 1.2 µm to 2.1 µm in the lateral direction and 1.1 µm to 4.5 µm in the axial direction for imaging positions in the range of 0 to 10 µm above the front focal plane, respectively.

 

Fig. 3. Raw images captured by TFMI-LFM system using AEC with FWHM of 21.6 µm for 100-nm florescent beads located (a) on, (b) 5 µm above, and (c) 10 µm above the front focal plane of the objective lens. (d)-(f) Reconstructed images of beads located at distances of 0 to 10 µm above the focal plane with reconstruction layer intervals of 1 µm. (g) and (h) Intensity profiles of three colored squares in Figs. 3(d)–3(f) in lateral and axial directions, respectively. Note that the blue, red and orange profiles correspond to 100-nm florescent beads located on, 5 µm above, and 10 µm above the front focal plane, respectively (Visualization 1 shows 3D dynamic observation for the Brownian motion of a 2-µm florescent bead in water with a volumetric imaging rate of 100 Hz).

Download Full Size | PDF

3.3 Scanless volumetric bioimaging via TFMI-LFM

The scanless volumetric bioimaging capability of the TFMI-LFM system was evaluated by imaging the lobe of a drosophila MB stained with GFP genetic fluorescent marker (OK-107). Figure 4(a) shows the raw image obtained from the TFMI-LFM system using an AEC with a FWHM of 21.6 µm. (Note that the small circular regions of the image correspond to the individual lenses in the MLA.) Figs. 4(b)–4(d) shows the reconstructed images corresponding to reconstructions depths 5 µm below, on, and 5 µm above the front focal plane of the TFMI-LFM system, respectively. Figure 4(e) shows the volumetric image obtained by stacking the 21 reconstructed images obtained over the depth range of 10 µm below the front focal plane to 10 µm above the front focal plane at intervals of 1 µm. (Note that each layer is reconstructed using a different color.) After the TFMI-LFM imaging process, the FWHM of the AEC was switched to 2.9 µm and the drosophila brain sample was imaged sequentially by the TF-MPEM system by axially moving the stage from a position 10 µm below the front focal plane of the TFMI-LFM system to a position 10 µm above the focal plane in steps of 1 µm. Figures 4(f)–4(h) show the reconstructed images corresponding to depths 5 µm below, on, and 5 µm above the front focal plane of the TFMI-LFM system, respectively. Figure 4(i) shows the volumetric image obtained by stacking the 21 TF-MPEM images obtained over the scanning depth range using a different color for each layer. To provide the corresponding high-resolution images as reference, a same image stack was captured by using a point-scanning MPEM system, as shown in the last row of Fig. 4. Figure 4(j) shows the whole single lobe of the drosophila MB, and the red square indicates the zoom-in imaging region of the lobe for the rest of Fig. 4. Figures 4(k)–4(m) present the point-scanning MPEM images at the same corresponding depths. Also, Fig. 4(n) shows the volumetric image by stacking the 21 point-scanning MPEM images obtained over the scanning depth range. Compared with the TF-MPEM and point-scanning MPEM images, the TFMI-LFM images present the lowest spatial resolution. However, the main structural features of the lobe can still be distinguished. Moreover, similar shapes and patterns are observed in Figs. 4(e), 4(i), and 4(n), respectively, and a similar 3D morphology is also observed. The exposure time of the TFMI-LFM system is 0.1 s, which is just the same time for getting single layer via the TF-MPEM system. However, the whole volume rate of the TFMI-LFM system, based on one-snapshot recording, is 21 times faster than that of the TF-MPEM system (i.e., to reconstruct 21 layers in the volume) and depends only on a sufficient fluorescent signal and sensitive image detector. Basically, the point-scanning system has better image contrast due to less crosstalk among pixels compared to the widefield recording methods such as the TF. Similarly, the LF records the whole axial information in one snapshot, and hence this 3D recording technique has more crosstalk from different pixels and layers. Furthermore, due to the limited NA of the MLA, the spatial resolution is gradually degraded away from the original focal plane [36]. The exposure time of the TFMI-LFM system for each is around twice that of the TF-MPEM system. However, the whole volume rate of the TFMI-LFM system, based on one-snapshot recording, is faster than that of the TF-MPEM system and depends only on a sufficient fluorescent signal and sensitive image detector.

 

Fig. 4. (a) Raw lobe image captured by TFMI-LFM system with 21.6-µm AEC, and corresponding reconstructed images at depths of: (b) 5 µm below, (c) on, and (d) 5 µm above the front focal plane. (e) Stacked image based on 21 reconstructed images acquired at 1-µm intervals over the depth range of 10 µm below to 10 µm above the front focal plane. Lobe images captured by TF-MPEM system with 2.9-µm AEC at depths of: (f) 5 µm below, (g) on, and (h) 5 µm above the front focal plane. (i) Stacked image based on 21 TF-MPEM images acquired at 1-µm intervals over the depth range of 10 µm below to 10 µm above the front focal plane. (j) Whole single lobe image captured by point-scanning MPEM system, and red square as the zoom-in imaging region of lobe. The high-resolution point-scanning images at depths of: (k) 5 µm below, (l) on, and (m) 5 µm above the front focal plane. (n) Stacked image based on 21 point-scanning MPEM images acquired at 1-µm intervals over the depth range of 10 µm below to 10 µm above the front focal plane (see Visualization 2 for the rendering process of the volumetric images using the 21 TFMI-LFM, TF-MPEM, and point-scanning MPEM images, respectively). Scale bar: 10 µm.

Download Full Size | PDF

4. Conclusions

This study has developed a TFMI system, in which the volumetric excitation effect is adjusted by means of a switchable lens pair with different magnifications placed in front of the objective. The TFMI system has been integrated with a LFM. The results have shown that the AEC of the TFMI-LFM system has a FWHM around seven times larger than that of a conventional TF-MPEM system (i.e., 21.6 µm vs. 2.9 µm). The TFMI-LFM imaging results have been reconstructed into 3D volumetric images using an iterative phase-space deconvolution algorithm. The imaging results obtained for the 100-nm fluorescent beads have shown that the TFMI-LFM system has lateral and axial resolutions of 1.2 µm and 1.1 µm, respectively, at the front focal plane. The feasibility of the proposed system for scanless volumetric bioimaging has been demonstrated by imaging the lobe structure of the drosophila MB. It has been shown that the imaging results have a poorer spatial resolution than those obtained using a TF-MPEM system via axial scanning to render a volume image. However, a similar native mesh characteristic is observed in the images captured by both systems. The imaging speed is greatly improved as the TFMI-LFM based on one-snapshot volume recording. Thus, overall, the results show that the TFMI-LFM system with multiphoton epifluorescence setup provides a promising approach for tissue-level imaging with a high-volume rate and micron-scale resolution in an easily-implemented way.