Fast 3D microscopic imaging methods have been playing a crucial role in many biological studies. In this Letter, we present a revolutionary way to design and build a two-photon excitation (TPE) microscope for 3D and random-access imaging based on a digital micromirror device (DMD), achieving a scanning speed of 22.7 kHz. When pairing with a objective lens, the maximum scanning range in the , , axes are 103, 206, 524 μm, respectively. The axial and lateral scanning resolution (i.e., minimum step size) are 270 nm and 130 nm, respectively. In the system, the focal point of the femtosecond laser can be arbitrarily positioned to any random point in space by switching the binary holograms stored in the DMD. Parametric models are derived to deterministically link the DMD parameters (i.e., pixel size and aperture) with the scanner characteristics, i.e., scan range and minimum step size, in each axis. In the experiments, we demonstrate conventional raster scanning, scanning along arbitrarily programmed surfaces, and random-access scanning on a pollen grain sample via the DMD-based TPE system. We also perform experiments to demonstrate the unique capability of selective optical stimulation, where selected locations within the specimen are photobleached by extending the laser dwell time. With its versatility and high scanning rate, the TPE microscope may find important applications in brain research, realizing in vivo random-access imaging and optical stimulation with tens of microseconds temporal resolution.
© 2017 Optical Society of America
The study of brain and neuronal activities calls for new imaging technologies, i.e., random-access microscopy, that can realize 3D imaging with high spatial and temporal resolution. Random-access microscopy was first developed to capture high-speed biological events on the scale of a few milliseconds, e.g., spiking of neural circuits, that typical raster scanning systems cannot achieve [1,2]. A typical 3D random-access microscope employs four acousto-optic deflectors (AODs) to arbitrarily access any point within the work volume at 10 s–100 kHz speed [3,4]. Due to the superior speed, a random-access microscope can maximize the scanning time at selected regions, resulting in better signal-to-noise ratios at much higher frame rates than raster scanning systems. Although the AOD-based scanning methods are fast (tens of kilohertz), the resolution is usually compromised as the focal point scans away from the original focal plane ; in addition, the large longitudinal size of the point spread function (PSF) ()  makes it unsuitable for volumetric scanning. A variety of methods have been developed to overcome these issues, e.g., separation of the lateral and axial scanning paths to minimize interferences among AODs , and application of high-bandwidth custom-built AODs in combination with high-performance control systems . These systems are complex and expensive, and require the four AODs to be synchronized; additional prism pairs are often required to compensate for the severe temporal dispersion from the AODs, making them difficult to use and maintain. A compact, low-cost, and versatile random-access imaging system that has comparable resolution to raster-scanning systems has yet to be developed.
In this Letter, we present a revolutionary way to design and build a two-photon excitation (TPE) microscope for random-access and 3D imaging as well as selective optical stimulation based on a single digital micromirror device (DMD). The TPE microscope achieves a scanning rate of 22.7 kHz, where the maximum scanning range in the , , axes are 103, 206, 524 μm, respectively. The axial and lateral scanning resolutions (i.e., minimum step size) are 270 nm and 130 nm, respectively. Previously, we have demonstrated high-speed femtosecond laser beam shaping and axial scanning using a DMD and binary holography [6,7]. This work is built upon the previous work, which extends the scanning capability to include the entire 3D space, achieving simultaneous axial and lateral scanning via superposing and rapidly modulating the designed binary holograms on the DMD. As every hologram corresponds to a specific point in space, the DMD scanner can move the focal point to any random location within its work volume range at equal speed, thereby realizing random-access imaging.
The optical configuration of the DMD-based TPE microscope is presented in Fig. 1. The laser source is a Ti:sapphire femtosecond laser (Chameleon Ultra II, Coherent) with a repetition rate of 80 MHz and an average power of 3.3 W; the laser is operated at a central wavelength of 800 nm with a pulse duration of 200 fs and a beam diameter of 2.5 mm. First, the input laser beam is expanded by L1 and L2 to fill the DMD aperture (DLP 4100 XGA, Texas Instruments). As the DMD functions both as a programmable binary hologram and a blazed grating, it introduces negative angular dispersion to the laser beam. To remove the angular dispersion, a blazed transmission grating (1200 lines/mm, Lightsmyth) and a mirror (M1) are included in the light path to generate positive angular dispersion. After the DMD, the dispersion-free laser beam is guided to an infinity-corrected objective lens (CFI Apo LWD N.A. 1.15, Nikon) via a telescope consisting of an achromatic lens L5 and a tube lens L6. An iris diaphragm is placed at the back focal plane of L5 to spatially reject all diffracted beams except the st-order diffraction of the hologram. Between the objective lens and L6, a long-pass dichroic mirror is installed to guide the emissions from the specimen to a photomultiplier tube (PMT) (R10699 and C7950, Hamamatsu), which is synchronized to the DMD trigger signals. Lastly, a data acquisition card collects and processes the PMT signals to form 3D images. These operations are fully automated by a custom-developed LabVIEW program.
Axial scanning along the optical axis (i.e., axis) is realized by applying binary holograms of spherical wavefronts of increasing or decreasing focal lengths :1) are generated based on Lee holography [8,9], and can be calculated as
Lateral scanning can be realized by varying the tilted phase term, i.e., in Eq. (2), where determines the tilt angle () of the fringe patterns, and determines the period of the fringes. This can be better understood in the frequency domain. The spatial frequency of the fringes in the and axes, i.e., and , respectively, can be found by performing a Fourier transform of the tilted phase term, i.e., . Figure 2 illustrates the lateral scanning process. To scan the focal point along the axis in the focal plane of L5, one can vary the value of that controls the separation distance (or angle) between the st and 0th diffraction order. Note that except the st order, all other diffracted beams are blocked by the spatial filter. Likewise, lateral scanning along the axis can be realized by adjusting the value of , and is independent of the scanning control in other axes. When selecting the operation parameters, it is worthwhile to note that small or high values of and , respectively, should be avoided, as they may cause the diffracted beams to overlap in space or the tilted phase term to exceed the range of the inequality in Eq. (2), resulting in no modulation. Combined with the aforementioned axial scanning method, 3D random-access scanning can be realized by a single DMD. As the DMD scanner is capable of scanning arbitrary paths in space, all conventional scanning strategies, e.g., raster, spiral, or Lissajous scanning trajectories, can be easily implemented on the DMD platform. In addition, the dwell time at each point on the scanning path can also be arbitrarily controlled, realizing selective optical stimulation.
In this section, we present parametric models that deterministically link the DMD parameters (i.e., pixel size and aperture) with the characteristics of the DMD scanner, including the scanning range and minimum step size. First, we consider the maximum scan range in the direction. Let , where is the number of pixels per grating period in the direction, and is the DMD pixel size. The maximum spatial frequency variation can be expressed as , where and are the minimum and maximum number of pixels per grating period in the direction, respectively. For our DMD, . Considering that the incident angle of the laser beam is 45°, the effective pitch () becomes 19.35 μm. Based on the geometric relationship in Fig. 1, the diffraction angle of the st order is proportional to the lateral shift on the focal plane of the objective lens, i.e., , where is the lateral shift in the sample plane and is the magnification of the objective lens. From the grating equation , we can obtain the lateral scan range in the direction:
Next, we derive the lateral scan range in the direction. Let ; is the number of pixels per grating period in the direction. Since lateral scanning along the axis is not constrained by a spatial filter, the maximum spatial frequency variation is twice than that in the axis, and can be expressed as , where is the minimum number of pixels per grating period in the direction. Following the same procedure, the lateral scan range is derived as3. Based on Eqs. (3)–(5), a larger scan range can also be achieved by using a DMD of more pixels or smaller pixel sizes. For example, if the DLP 6500 is used, the work volume will be increased by a factor of .
Next, we consider the minimum step size in the lateral directions. Here, the minimum step size is defined as the smallest variation of (i.e., ) that will result in a consistent step, assuming the DMD aperture is fully filled by the laser beam. can be found when the titled phase varies at half the DMD aperture (i.e., and ). Substituting this requirement in Eq. (2), we arrive at
Rearranging Eq. (6), we have . Substituting this result in Eq. (3), the minimum step size for -axis scanning () is found and expressed in Eq. (7). Note that the expression for the minimum step size in the direction () is the same as Eq. (7),
Let and , is calculated to be for our system. As reported in , the axial scanning resolution of the DMD scanner is . These results indicate that the DMD-based TPE microscope has suitable scanning resolution for performing continuous scanning both in the lateral and axial directions. The minimum step size can be further reduced by using DMDs of more pixels, as indicated in Eq. (7).
To verify the performance of the DMD-based TPE microscope, we performed 3D imaging experiments on a pollen grain sample (, mixed pollen grains, Carolina Biological Supply). In the first experiment, the DMD is programmed to perform raster scanning, where the volume image of a single grain of pollen, sized , is obtained. Eighty thousand binary holograms are automatically generated and loaded to the random-access memory (RAM) of the DMD controller for performing the task. The scanning is performed at 22.7 kHz, and the entire volume imaging process is completed within 3.6 s. Figures 4(a)–4(h) present the results of eight cross-sectional layers of the pollen, where individual spikes on the grain can be clearly visualized. Each layer has pixels and is 5 μm apart. From the results, one can conclude that the imaging resolution and optical cross-sectioning capability of the DMD-based system is comparable to the conventional laser scanning TPE microscopes.
To demonstrate the capability of scanning arbitrary paths in space, we scan two different pollen grains on spherical and sinusoidal surfaces. The results are presented in Fig. 5. (See Visualization 1 for video demonstrations.) For the spherical surface, the scanning parameters should satisfy the surface function, i.e., , where and are constants; , , and are proportional to the , , and , respectively. The sinusoidal surface profile is obtained via a similar procedure. The scanning is performed at 22.7 kHz, resulting in a frame rate of .
Next we demonstrate random-access imaging on our DMD-based TPE microscope. Six distant points “on,” “inside,” or “outside” the pollen grain are selected among the eight layers for fast random-access imaging. The selected points are labeled by different colors, as shown in Fig. 6(a). Next, binary holograms associated with the six selected points are generated and loaded to the DMD controller for continuous scanning. Figure 6(b) presents the scanning results, where fluorescence intensity variations of the six points are recorded over time. The color bar in Fig. 6(b) associates the fluorescence data to specific points in Fig. 6(a). From the time scale in Fig. 6(b), where the time interval is 44 μs, we can confirm that the DMD scanner can perform random-access scanning at the DMD pattern rate, i.e., 22.7 kHz.
In the last experiment, we demonstrate 3D-resolved random-access optical stimulation. This can be realized by simply extending the dwell time of the laser focus at any selected point in space; for example, a dwell time of 1 ms allows 80,000 laser pulses to impinge upon the specimen, providing sufficient power and speed to stimulate, blaze, or even cut individual cells or tissues. Figures 7(a)–7(c) present the selective optical stimulation results, where the selected seven spikes on the pollen volume image (shown in Fig. 4) are exposed to laser pulses for a total period of 0, 3, and 15 s, respectively, at a low power level, i.e., 30 mW. From the results, one can clearly observe the effect of photobleaching, where the spikes no longer emit fluorescence signals and the adjacent areas remain unaffected.
For efficiency, the overall efficiency is measured to be ( for the input beam, for the output beam after the objective lens), where most energy is lost from the diffraction of the DMD . Although the efficiency is relatively low, it may be easily compensated for by increasing the illumination power of the femtosecond lasers.
In conclusion, we have presented the design, modeling, and characterization of a new TPE microscope based on a single DMD that realizes random-access and 3D imaging as well as selective optical stimulation with a scanning rate of 22.7 kHz; a maximum scanning range of 103, 206, 524 μm in the , , axes, respectively; and scanning resolution (i.e., minimum step size) of 270 nm and 130 nm in the axial and lateral directions, respectively. Parametric models have been developed to predict the maximum scan range and minimum step size of the DMD scanner in each axis with experimental verification. The model also indicates that the performance of the TPE microscope may be further improved by employing DMDs of more pixels or smaller pixel sizes. In the experiments, raster-scanning, random-access scanning, and scanning over arbitrarily defined surfaces (i.e., spherical and sinusoidal surfaces) have been performed on pollen grain samples to demonstrate that the TPE microscope can operate at the DMD pattern rate (22.7 kHz) with a resolution comparable to conventional laser scanning TPE microscopes. Seven arbitrarily selected points “on” or “inside” the pollen grain have been photobleached by extending the dwell time of the laser focus, showing potential application of selective optical stimulation or precise laser surgery on cells or tissues. More importantly, many existing optical methods can be simultaneously implemented on the TPE microscope via a single DMD, e.g., wavefront correction or compressive sensing, by superposing new hologram programs of different purposes to the scanning holograms. With its versatility and high scanning rate, the new DMD-based TPE microscope may generate significant impact to the field of biomedical imaging.
General Research Fund (GRF); HKSAR Research Grants Council, University Grants Committee (RGC, UGC) (CUHK 439813, CUHK 14202815).
1. A. Bullen, S. Patel, and P. Saggau, Biophys. J. 73, 477 (1997). [CrossRef]
2. V. Iyer, T. Hoogland, and P. Saggau, J. Neurophysiol. 95, 535 (2006). [CrossRef]
3. G. Reddy and P. Saggau, J. Biomed. Opt. 10, 064038 (2005). [CrossRef]
4. G. Katona, G. Szalay, P. Maak, A. Kaszas, M. Veress, D. Hillier, B. Chiovini, E. Vizi, B. Roska, and B. Rozsa, Nat. Methods 9, 201 (2012). [CrossRef]
5. K. Nadella, H. Ros, C. Baragli, V. Griffiths, G. Konstantinou, T. Koimtzis, G. Evans, P. Kirkby, and R. Silver, Nat. Methods 13, 1001 (2016). [CrossRef]
6. J. Cheng, C. Gu, D. Zhang, and S. Chen, Opt. Lett. 40, 4875 (2015). [CrossRef]
7. J. Cheng, C. Gu, D. Zhang, D. Wang, and S. Chen, Opt. Lett. 41, 1451 (2016). [CrossRef]
8. W. Lee, Appl. Opt. 13, 1677 (1974). [CrossRef]
9. O. Bryngdahl and W. Lee, Appl. Opt. 15, 183 (1976). [CrossRef]
10. F. Fahrbach, F. Voigt, B. Schmid, F. Helmchen, and J. Huisken, Opt. Express 21, 21010 (2013). [CrossRef]