Open Access
Add to BibSonomy
Abstract
Fast 3D volume imaging methods have been playing increasingly important roles in biological studies. In this article, we present the design and characterization of a multi-focus line-scanning two-photon microscope. Specifically, a digital micromirror device (DMD) is employed to generate a randomly distributed focus array on a plane (i.e., x-z) via binary holography. Next, a galvanometric mirror scans the focus array in a direction normal to the plane (i.e., y-axis) over the imaging volume. For sparse samples, e.g., neural networks in a brain, 1-3 foci are used together with compressive sensing algorithm to achieve a volume imaging rate of 15.5 volumes/sec over 77 × 120 × 40 µm3. High-resolution optical cross-sectional images on selected planes and regions can be generated by sequentially scanning the laser focus generated on the x-z plane with good imaging speeds (e.g., 107 frames/sec over 80 × 120 × 40 µm3). In the experiments, microbeads, pollens, and mouse brain slices have been imaged to characterize the point spread function and volume image rate and quality at different sampling ratios. The results show that the multi-focus line-scanning microscope presents a fast and versatile 3D imaging platform for deep tissue imaging and dynamic live animal studies.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
3D two-photon fluorescence microscopy has been widely used to study functional and structural information in biological specimens with subcellular resolution and powerful optical cross-sectioning capability [1]. However, common two-photon microscope systems acquire images by serially scanning the laser focus and thus cannot simultaneously investigate different dynamic biological events in a 3D space. As such, it is imperative to develop high-speed volume imaging methods for two-photon microscopy. One way to parallelize the scanning process is to split a laser focus into multiple foci at different planes to increase the volume image speed [2–5]. To minimize the cross-talk among the split foci, spatial and temporal multiplexing methods have been developed to separate the laser foci both spatially and temporally [6–9] at the expense of requiring a low repetition rate and high-power femtosecond light source.
In contrast to point scanning, line-scanning and plane-scanning offers orders of magnitude speed enhancement, which has been widely implemented in single-photon microscopes, e.g., light sheet microscope [10–13]. To implement plane-illumination in a two-photon system, a light sheet can be generated by two cylindrical lenses or alternatively by scanning a laser beam [14–18]. Two-photon excitation light sheet microscopy can also be generated by spatially and temporally focusing a femtosecond laser [19] to achieve high-speed volume imaging at the expense of compromised axial resolution, which can be alleviated by applying post image processing algorithms, such as structured light illumination [20,21]. Axial scanning in such systems are typically realized by mechanical scanning, adjusting group velocity dispersions of the laser spectrum (e.g., via a pair of prisms), or using an electrically tunable lens [22–25]. Notably, in general wide field imaging systems have a lower signal-to-noise ratio (SNR) in thick biological specimens in comparison to point-scanning systems owing to more a pronounced scattering effect [26].
Considering many biological specimens are intrinsically sparse, such as neural networks in a brain, compressive sensing (CS) algorithms can be applied to increase the temporal resolution of an imaging system [27–29]. To apply CS in a fluorescent microscope, the light source needs to be modified to provide programmable and sparse illumination patterns, which is typically realized by combining a laser with a spatial light modulator. The first CS-based laser-scanning confocal microscope was developed in 2016 [30], where 2D images were reconstructed from sub-sampled signals by a convolution and CS algorithm. As the galvo-mirror only scans the laser in the focal plane, 3D images can only be obtained by stacking the reconstructed 2D images. To directly reconstruct a 3D image, Wen et al. applied a random-access DMD optical engine to generate 1-10 laser foci in the imaging field in combination with CS to increase the imaging speed of a two-photon microscopy with the flexibility to selectively image any region of interests at a sampling ratio of 20% [31]. However, since a serial scanning system is intrinsically slower than a line-scanning system, the resulting imaging speed is still limited to 9.1 frames/second (fps).
In this work, we present a CS-based DMD-based multi-focus line-scanning microscopy to realize high-speed volume imaging with close to diffraction limit resolution. Specifically, a DMD is employed to generate a randomly distributed focus array on a plane via binary holography. Next, a resonant galvanometric mirror scans the focus array in a direction normal to the plane over the imaging volume, achieving a single-point scanning rate of 107 fps (512 × 148 voxels) with diffraction-limited resolution and a multi-focus volume imaging rate of 15.5 volume/sec (vps) over volume of 77 × 120 × 40 µm3 with the application of CS algorithm.
2. Materials and methods
2.1 System design
Figure 1(a) presents the optical configuration of the multi-focus line-scanning two-photon microscope. In the system, the light source is an 80 MHz Ti-Sapphire femtosecond laser (Chameleon Ultra II, Coherent) with a pulse width of 140 fs and a central wavelength of 800 nm. The wavelength is tunable between 680 nm and 1080 nm. (The average power at 800 nm is 3.5 W.) To pre-compensate the angular dispersion introduced by the DMD, we first direct the laser to a blazed Grating: 1208 grooves/mm). Notably, this design enables broadband dispersion compensation (750-1050 nm) for the DMD [32]. Next, the laser beam is expanded to 10-mm diameter to fully fill the DMD aperture via lenses L1 (fL1 = 100 mm) and L2 (fL2 = 250 mm). The DMD (DLP 4100) modulates the incoming laser wavefront via binary holography [33,34] to generate the designed laser focus array on the x-z plane. After the DMD, the first-order diffracted beam is selected by a spatial filter (SF) and collimated by a telescope system, consisting of L3 (fL3 = 400 mm) and L4 (fL4 = 150 mm). A resonant galvanometric mirror (CRS 8KHz, Cambridge Technology) synchronized with the DMD then scans the hologram-generated focus array in the y-direction. (Note the galvo operates in the bidirectional scanning mode with a line-scanning frequency of 15,837 Hz, which is comparable with DMD pattern rate of 22.7 kHz.) Finally, the line-array is relayed to the focal region of the objective lens (Fluor $40 \times $/0.80W DIC, Nikon) via lenses L5 (fL5 = 54 mm) and L6 (fL6 = 200 mm) and a high reflectivity mirror (M) as well as a long-pass dichroic mirror (DM, ZT670rdc, Chroma). The emissions from samples are collected by the objective lens, DM, and PMT. As illustrated in Fig. 1(b), by quickly switching the designed holograms on the DMD, the two-photon system enables different imaging modes for in vivo biological studies, including the (1) conventional single-focus raster-scanning mode (left), (2) single-focus programmable imaging plane mode (middle), and (3) multi-focus line-scanning mode (in combination with CS to reconstruct volume images) (right). Imaging results of each mode are presented in Section 3.
Fig. 1. Optical configuration and principle of multi-focus line-scanning two-photon microscopy. (a) Optical setup. DMD: digital micromirror device; SF: spatial filter; M: mirror; DM: dichroic mirror; PMT: photomultiplier tube; Grating: 1208 grooves/mm; and L1-L6: lenses. VP1: viewpoint 1, VP2: viewpoint 2. (b) Scanning strategies, including (1) layer-by-layer raster-scanning (left), (2) programmable imaging plane (middle), and (3) multi-focus line-scanning (right). The color bar indicates time.
2.2 Reconstruction algorithm for multi-focus line-scanning
In the imaging process, the volume to be captured is pixelated into a grid of discrete elements. The dimension of the grid is denoted as Nx, Ny and Nz, representing the number of pixels in the x, y, and z directions, respectively. For the first and second imaging modes, each hologram generates one laser focus in the x-z plane, followed by fast line-scanning in the y-direction; this process repeats until the all hologram generated points on the x-z plane are scanned. As the relationship between focus position and collected emission is explicit, 2D or 3D images can be directly generated by registering the emission signals to its corresponding spatial coordinates. Although such serial scanning approach suffers from limited imaging rate, their results can serve as good references to compare with CS reconstructed images. In addition, as the DMD can flexibly generate scanning paths in the x-z plane, some scanning time may be saved by only scanning the regions of interest.
For multi-focus line-scanning, as one hologram encodes multiple laser foci, emissions from different probed locations are collected by a single PMT. To decode the recorded intensities at different locations, CS algorithm can be applied for image reconstruction. To ensure good image reconstruction, M sets of focus array will be used. (M / (N x $\times N_{z} $)= sampling ratio, i.e., RCS.) In each focus array, 2-3 laser foci are generated at randomly selected positions in the x-z plane. At each selectively excited x-z plane, the emissions are recorded by the PMT, which repeats Ny times at 15,837 Hz to complete the raw data acquisition for CS reconstruction. Accordingly, the total data acquisition time is M/15,837 ≈ 0.06-0.32 sec (i.e., volume imaging rate ≈ 3-15 vps), depending on the sampling ratio. In other words, each x-z plane measurement corresponds to a single y-direction sweeping process, and all recorded x-z slices can be described by a measurement matrix A. Accordingly, the relationship between the vectorized x-z slice (x) and recorded signals (b) can be expressed as:
Figure 2 illustrates the workflow of CS-based 3D image reconstruction. Step 1 presents the data acquisition process, which is described in the last paragraph. In Step 2, the recorded intensity data ([b1,:, b2,:, …, bM,:] ∈ RM×Ny) are reorganized to [b:,1, b:,2, …, b:,Ny]T, representing Ny vectorized x-z slices that has a dimension of M × 1 ready for CS reconstruction. Notably, as x-z slice is independent from others, all data can be processed in parallel to save computation time. In Step 3, CS algorithm is applied to parallelly process Ny x-z slices, which is mathematically expressed as:
Fig. 2. Working principle for multi-focus line-scanning with CS. Step 1: scan M different hologram-generated focus patterns to obtain the raw data. Step 2: reorder the collected M datasets to obtain ${N_y}$ slices in the y-direction. Step 3: apply CS algorithm on each slice to obtain a 3D image.
3. Results
We have devised and performed imaging experiments to demonstrate the new capabilities of the multi-focus line-scanning two-photon system. First, we characterized the lateral and axial resolution of the two-photon system using 200-nm fluorescent beads (F8811, Thermo Fisher Scientific). In the experiments, the samples were axially scanned over a 9-µm range with a step size of 300 nm. The lateral and axial point spread function (PSF) are presented in Fig. 3(a) and 3(b) respectively, where where the blue/red squares and solid lines represent the normalized intensity data and fitted Gaussian curves, respectively. Accordingly, the full width at half maximum (FWHM) of the two-photon system is 0.60 µm, 0.66 µm in the lateral directions; and 4.03 µm in the axial direction. This represents the imaging resolution of the first and second imaging modes and the performance limit of the CS-reconstructed 3D images.
Fig. 3. Characterization of the lateral (a) and axial (b) PSF via 200-nm fluorescent beads. “The blue/red squares and solid lines represent the normalized intensity data and fitted Gaussian curves. Scale bar: 1 µm
Next, we demonstrate the flexibility of the DMD-enabled serial scanning process (i.e., the first and second imaging modes) on pollen samples and mouse brain slices. The results are presented in Fig. 4. In the experiments, the lateral and axial pixel sizes were set as 300 nm and 1 µm respectively. In Fig. 4(a)–4(c), we selected a small region of interest (60 × 60 × 30 µm3) on the pollen sample, where the red patterns in the bottom left corner of each sub-figure represent the hologram-generated scanning paths. The results show well-defined optical cross-sectional images along different directions (Fig. 4(a)) and planes (Fig. 4(b)). Figure 4(c) presents a sinusoidally shaped image volume, which are results commercial two-photon systems cannot achieve. Figure 4(d) presents imaging results of the mouse brain slice. Again, a small region of interests (80 × 120 × 40 µm3), which contains two neuron cells separated by 9 µm axially, was selected to demonstrate the flexibility of the DMD scanner. Figure 4(e) shows a zoom-in view of the yellow box labeled in Fig. 4(d). Importantly, because of the small region of interests and high galvo scanning rate (15,837 Hz), imaging results in Fig. 4(a)–4(c) and Fig. 4(e) are obtained with an imaging rate of 61, 26, 6, and 107 fps, respectively. These results demonstrate the flexibility of the DMD scanner and the power of programmable imaging planes.
Fig. 4. Demonstration of hologram-based selective serial scanning over a small region of interests. (a)-(c) optical cross-sectional imaging of pollen samples in different planes; (d) 3D imaging of neuron cells (80 × 120 × 40 µm3) in a mouse brain slice; (e) zoom-in view of yellow box in (d). Scale bar: 15 µm
Lastly, we demonstrate high-throughput 3D imaging of brain slices of a transgenic mouse that express green fluorescent protein (GFP) via the CS-based multi-focus line-scanning two-photon microscope. The results are presented in Fig. 5(a). In the experiments, we first selected a small target volume of 77 × 120 × 40 µm3 (corresponding to 256 × 400 × 40 voxels) at 100 µm beneath the sample. To investigate the relation between imaging quality and speed, we performed the imaging experiments with a sampling ratios (Rcs) of 10%, 30%, and 50%, respectively, which correspond to a volume imaging rate of 15.5, 5.2, and 3.1 vps. At each sampling ratio, 1-3 laser foci were used to excite the sample. (Note that based on our previous study [31], good reconstruction results can be achieved when the number of foci is below 5.) For comparison, a reference high-resolution 3D image was acquired via the raster-scanning mode (Fig. 5(a)) at 1.6 vps (i.e., 62 fps for 40 layers). Figure 5(b) presents enlarged views of the selected regions in Fig. 5(a) for comparison.
Fig. 5. 3D imaging results of the mouse brain slice with different sampling ratios and number of foci, where the imaging depth is color-labeled. (a) The top row shows the reference image from the serial scanning mode. The second, third and fourth rows present CS-reconstructed images with a sampling ratio of 10%, 30%, and 50%, respectively, which correspond to an imaging rate of 15.5, 5.2, and 3.1 vps. Scale bar: 30 µm (b) Zoom-in view of the yellow box in the reference image in (a) as well as the zoom-in views of different sampling ratios and number of foci at the same position for comparison. Scale bar: 5 µm. (c) Relative error and SSIM plot of reconstructed images under different sampling ratios and number of foci.
From the results, we may find that a higher sampling ratio always yields better reconstructed images. A sampling ratio of 10% enables high-speed 3D imaging (15.5 vps) at the expense of slightly compromised image quality, as shown in Fig. 5(a), where most structural information of the neural network and neuron cells were preserved. When the sampling ratio is of 30% or above, reconstruction errors and artifacts are visually negligible. To quantitatively assess the reconstructed images, the relative error ($\left \| x-\bar{x} \right \|_{2}/\left \| x \right \| _{2} $) and structural similarity index (SSIM) of each reconstructed image was calculated, as shown in Fig. 5(c). The results show that that the image quality enhances with increasing number of laser foci for all different sampling ratios, which is consistent with our observation in Fig. 5(b). On the other hand, at low sampling ratios and small number of foci, streak artefacts can be observed in many reconstructed images. This is because the image is scanned pixel-by-pixel in the y-direction, while reconstructed via the CS algorithm in the x-z plane.
Overall, we have experimentally verified that the multi-focus line-scanning method with CS can improve the imaging rate by a factor of 10 with slightly compromised image resolution, which is superior to many previously reported methods [35,36]. This suggests our method may find many imaging applications that demand speed and resolution. Examples include imaging of neuronal activities in vivo [37] in brain research or developmental biology [36].
4. Conclusion
In conclusion, we have presented the design and characterization of a CS-based multi-focus line-scanning two-photon scanning microscopy that combines a holography-based DMD scanner and resonant galvanometric mirror. The system provides three operation modes, including the (1) serial scanning mode, (2) 3D programmable imaging plane mode, and (3) CS-based multi-focus line-scanning mode. The first two modes can perform flexible 3D imaging at video rate with the capability to selectively image a small region of interest at higher speed (∼100 fps). Based on CS, the multi-focus line-scanning mode can substantially increase the imaging rate to 15.5 vps with slightly compromised resolution. In the experiments, we imaged different samples, including fluorescent beads, pollen grains, and mouse brain slices, to characterize the resolution, imaging rate, and different imaging functions and capabilities of each imaging mode. These results show that the multi-focus line-scanning microscope presents a fast and versatile 3D imaging solution to visualize high-speed dynamic biological events for different studies.
Funding
Research Grants Council, Collaborative Research Fund (C5031-22GF); Innovation and Technology Commission, Innovation Technology Fund (ITS/222/21FP); Hong Kong Center for Cerebro-cardiovascular Health Engineering (COCHE-1.6).
Acknowledgments
We thank Prof. Pak Kan Jacque Ip and Ms. Xinrong Li at the School of Biomedical Sciences, The Chinese University of Hong Kong for preparing mouse brain slices.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are available upon reasonable request.
Reference
1. W. Denk, J. H. W. Denk, J. H. Strickler, et al., “Two-photon laser scanning fluorescence microscopy,” Science 248(4951), 73–76 (1990). [CrossRef]
2. A. Egner and S. W. Hell, “Time multiplexing and parallelization in multifocal multiphoton microscopy,” J. Opt. Soc. Am. A 17(7), 1192–1201 (2000). [CrossRef]
3. W. Yang, J.-E. K. Miller, L. Carrillo-Reid, et al., “Simultaneous Multi-plane Imaging of Neural Circuits,” Neuron 89(2), 269–284 (2016). [CrossRef]
4. W.-C. A. Lee, H. Huang, G. Feng, et al., “Dynamic remodeling of dendritic arbors in GABAergic interneurons of adult visual cortex,” PLoS Biol. 4(2), e29 (2005). [CrossRef]
5. M. Buist and B. Squier, “Real time two-photon absorption microscopy using multi point excitation,” J. Microsc. 192(2), 217–226 (1998). [CrossRef]
6. D. R. Beaulieu, I. G. Davison, K. Kılıç, et al., “Simultaneous multiplane imaging with reverberation two-photon microscopy,” Nat. Methods 17(3), 283–286 (2020). [CrossRef]
7. W. Amir, R. Carriles, E. E. Hoover, et al., “Simultaneous imaging of multiple focal planes using a two-photon scanning microscope,” Opt. Lett. 32(12), 1731–1733 (2007). [CrossRef]
8. S. Xiao, J. T. Giblin, D. A. Boas, et al., “High-throughput deep tissue two-photon microscopy at kilohertz frame rates,” Optica 10(6), 763 (2023). [CrossRef]
9. T. Zhang, O. Hernandez, R. Chrapkiewicz, et al., “Kilohertz two-photon brain imaging in awake mice,” Nat. Methods 16(11), 1119–1122 (2019). [CrossRef]
10. J. Huisken, J. Swoger, F. Del Bene, et al., “Optical sectioning deep inside live embryos by selective plane illumination microscopy,” Science 305(5686), 1007–1009 (2004). [CrossRef]
11. Y. Chang, C. Wen, C. Gu, et al., “Synchronization-free light sheet microscopy based on a 2D phase mask,” Optica 4(9), 1030 (2017). [CrossRef]
12. Q. Zhong, A. Li, R. Jin, et al., “High-definition imaging using line-illumination modulation microscopy,” Nat. Methods 18(3), 309–315 (2021). [CrossRef]
13. P. Haslehurst, Z. Yang, K. Dholakia, et al., “Fast volume-scanning light sheet microscopy reveals transient neuronal events,” Biomed. Opt. Express 9(5), 2154–2167 (2018). [CrossRef]
14. J. Palero, S. I. Santos, D. Artigas, et al., “A simple scanless two-photon fluorescence microscope using selective plane illumination,” Opt. Express 18(8), 8491–9498 (2010). [CrossRef]
15. F. O. Fahrbach, V. Gurchenkov, K. Alessandri, et al., “Light-sheet microscopy in thick media using scanned Bessel beams and two-photon fluorescence excitation,” Opt. Express 21(11), 13824–13839 (2013). [CrossRef]
16. M. Kumar, S. Kishore, J. Nasenbeny, et al., “Integrated one- and two-photon scanned oblique plane illumination (SOPi) microscopy for rapid volumetric imaging,” Opt. Express 26(10), 13027–13041 (2018). [CrossRef]
17. T. V. Truong, W. Supatto, D. S. Koos, et al., “Deep and fast live imaging with two-photon scanned light-sheet microscopy,” Nat. Methods 8(9), 757–760 (2011). [CrossRef]
18. V. Voleti, K. B. Patel, W. Li, et al., “Real-time volumetric microscopy of in vivo dynamics and large-scale samples with SCAPE 2.0,” Nat. Methods 16(10), 1054–1062 (2019). [CrossRef]
19. D. Oron, E. Tal, and Y. Silberberg, “Scanningless depth-resolved microscopy,” Opt. Express 13(5), 1468–1476 (2005). [CrossRef]
20. Y. Meng, W. Lin, C. Li, et al., “Fast two-snapshot structured illumination for temporal focusing microscopy with enhanced axial resolution,” Opt. Express 25(19), 23109–23121 (2017). [CrossRef]
21. J. Chen, S. Gu, Y. Meng, et al., “Holography-based structured light illumination for temporal focusing microscopy,” Opt. Lett. 46(13), 3143–3146 (2021). [CrossRef]
22. M. E. Durst, G. Zhu, and C. Xu, “Simultaneous spatial and temporal focusing for axial scanning,” Opt. Express 14(25), 12243 (2006). [CrossRef]
23. Y. Ding, A. C. Aguilar, and C. Li, “Axial scanning with pulse shaping in temporal focusing two-photon microscopy for fast three-dimensional imaging,” Opt. Express 25(26), 33379 (2017). [CrossRef]
24. J. Jiang, D. Zhang, S. Walker, et al., “Fast 3-D temporal focusing microscopy using an electrically tunable lens,” Opt. Express 23(19), 24362–24368 (2015). [CrossRef]
25. D. Wang, Y. Meng, D. Chen, et al., “High-speed 3D Imaging based on Structured Illumination and Electrically Tunable Lens,” Chin. Opt. Lett. 15(9), 090004 (2017). [CrossRef]
26. E. Papagiakoumou, E. Ronzitti, and V. Emiliani, “Scanless two-photon excitation with temporal focusing,” Nat. Methods 17(6), 571–581 (2020). [CrossRef]
27. P. Ye, J. L. Paredes, G. R. Arce, et al., “Compressive confocal microscopy,” IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 429–432 (2009).
28. Y. Wu, P. Ye, I. O. Mirza, et al., “Experimental demonstration of an Optical-Sectioning Compressive Sensing Microscope (CSM),” Opt. Express 18(24), 24565–24578 (2010). [CrossRef]
29. V. Studer, J. Bobin, M. Chahid, et al., “Compressive fluorescence microscopy for biological and hyperspectral imaging,” Proc. Natl. Acad. Sci. U. S. A. 109(26), E1679 (2012). [CrossRef]
30. N. Pavillon and N. I. Smith, “Compressed sensing laser scanning microscopy,” Opt. Express 24(26), 30038–30052 (2016). [CrossRef]
31. C. Wen, M. Ren, F. Feng, et al., “Compressive sensing for fast 3-D and random-access two-photon microscopy,” Opt. Lett. 44(17), 4343–4346 (2019). [CrossRef]
32. D. Chen, B. Chen, Q. Shao, et al., “Broadband angular dispersion compensation for digital micromirror devices,” Opt. Lett. 47(3), 457–460 (2022). [CrossRef]
33. W.-H. Lee, “Binary Synthetic Holograms,” Appl. Opt. 13(7), 1677–1682 (1974). [CrossRef]
34. Q. Geng, C. Gu, J. Cheng, et al., “Digital micromirror device-based two-photon microscopy for three-dimensional and random-access imaging,” Optica 4(6), 674 (2017). [CrossRef]
35. H. He, C. Kong, X.-J. Tan, et al., “Depth-resolved volumetric two-photon microscopy based on dual Airy beam scanning,” Opt. Lett. 44(21), 5238–5241 (2019). [CrossRef]
36. Y. Gao, X. Xia, L. Liu, et al., “Axial gradient excitation accelerates volumetric imaging of two-photon microscopy,” Photonics Res. 10(3), 687 (2022). [CrossRef]
37. R. Lu, W. Sun, Y. Liang, et al., “Video-rate volumetric functional imaging of the brain at synaptic resolution,” Nat. Neurosci. 20(4), 620–628 (2017). [CrossRef]
