1. Introduction

Three-dimensional volume imaging is now an essential tool in biological, and other, application areas. Multiphoton scanning microscopy accomplishes this by scanning a point source through the sample volume, collecting the resultant fluorescence, and re-creating the image [1]. The resolution has a high dependence on the point-spread function (psf) of the scanned point source. Additionally, the signal levels in two-photon fluorescence have a strong nonlinear dependence upon the incident light intensity. Thus the ideal is to image the sample by use of a diffraction limited spot. However, refractive-index mismatches between the immersion media of high-NA objectives and the sample, in addition to the usually nonuniform samples themselves, impart aberrations to the wavefront and contribute to a deterioration of the psf. Other sources of aberrations include off-axis transmission through optical components, such as the objective and scan lenses while scanning, and imperfections in the optical elements throughout the optical train. The net effect of specimen induced aberrations is that while reducing the achievable resolution, the necessary laser power to achieve imaging increases with depth which can lead to nonreversible effects in biological samples. It also places greater demands in terms of output power on multiphoton imaging laser sources.

For a uniform sample with a refractive-index mismatch between the immersion fluid of the objective and the sample, the primary aberration is spherical which naturally increases with depth. Several methods for compensating for spherical aberration have been proposed and implemented. Varying the effective tube length [2], refractive index of the immersion fluid [3,4] and the pupil area [4] have been used with reasonable success. A range of high-NA objectives are available with correcting collars designed to correct for spherical aberration imparted by variations in the cover slip thickness but which can also be used to compensate for some specimen induced spherical aberration [4]. However, all of these are essentially static corrections and not easily changed during imaging.

Dynamic correction of aberrations imparted by an interposing media has been performed in military and astronomical applications for a number of years [5]. As this technology of adaptive optics has matured, the areas to which it can be applied have correspondingly broadened. In the area of microscopy, a liquid crystal spatial light modulator has been used to apply the phase conjugate of measured specimen induced aberrations to the imaging laser beam in a two-photon microscope [6]. Although a successful technique, the wavefront generator had an optical throughput of only 1 – 2% at 780nm. Another group has used a deformable mirror (DM) to compensate for specimen induced aberrations, in this case the sample was Coumarin in water [7]. As in the previous implementation, wavefront correction was applied and relayed to the objective through the scanning components. In the latter case, the laser beam had to be up-collimated to match the laser to the DM then down-collimated to match the entrance pupil of the microscope objective with a second 4f telescope. Placing the DM at a significant distance from the objective meant that when using a defocus bias, which was necessary to perform aberration correction at greater depths in the sample, a 50% reduction in the laser beam transmission occurred due to the over-filling of the objective. In confocal microscopy, a DM was also used for aberration correction [8]. In this embodiment, the specimen was mounted on a scanning stage to facilitate volume scanning with the laser beam being incident at 16 degrees on the DM before being sent to the objective.

The research described here was driven by a requirement to image optically aberrating biological samples having feature sizes in the order of several micrometers over an area of 32µm squared. Repetitive scans of the same volume were required without movement of the sample. Thus fast scan rates and a high photon efficiency in order to reduce bleaching effects and image acquisition times were necessary to effectively fulfil the requirements. Presented here is a nondescanned multiphoton microscope incorporating a deformable mirror and capable of scan rates of 0.3 to 4 frames per second with automatic optimization of the psf using intensity feedback from the two-photon induced fluorescence.

2. Experiment

The experimental layout is shown in Fig. 1. A 76MHz, 200fs commercial Ti:sapphire laser (Coherent Mira) was used as the excitation source of the two-photon induced fluorescence. 750nm was selected as the initial investigation wavelength for the simple expedient that it is visible. However, all components inserted in the optical train were achromatic to allow tunability over the full wavelength range of the Ti:Sapphire laser. Post scan-head implementation of the adaptive optic was chosen in order to produce a compact design and allow easy integration to the system. In addition, the short optical path length between the DM and the objective minimizes the changes in beam input diameter to the back aperture of the objective, thus reducing changes in transmission through the system as the mirror is deformed.

 

Fig. 1. Experimental apparatus.

Download Full Size | PDF

Due to the scanning mechanism of the MRC 600, there is a stationary spot on the output galvo on the axis running perpendicularly, and through, the center of the eyepiece lens. By appropriate adjustment of the eyepiece and scan lens in the beam-expanding telescope, an essentially stationary spot can be formed on the surface of the DM. The linearly polarized laser light from the eyepiece traverses the scan lens and in turn the polarizing beam splitter and the quarter wave plate before being reflected off the DM, back through the quarter wave plate and redirected by the polarizing beamsplitter towards the objective. A dichroic allows the laser light to pass and redirects the fluorescence to the PMT. This forms a compact unit with no unnecessary lossy components between the objective and the PMT. It will be noted that the trade-off for such a compact design is that there is a minimal amount of beam movement on the back aperture of the objective. For much larger scan angles than were involved in this study, additional optics would be necessary to create a second stationary spot on the back aperture of the objective. If required, a defocus bias can be applied to the wavefront by the appropriate movement of the scan lens. This can be used to compensate for the fact that the DM produces only concave surfaces and thus restore the parallelism of the scanning laser beam. Axial focussing was provided by mounting the objective in a closed loop piezoelectric displacement device (P-721.17) manufactured by Physik Instrumente.

The adaptive optic used was a 15mm diameter DM manufactured by OKO Technologies. The deformation of the silicon nitride membrane is controlled by 37 electrostatic actuators [9]. A 3.5µm deformation was measured using an interferometer set-up when using the maximum voltage from the amplifier circuit, a value inside that which could cause damage to the mirror. A key feature is that the laser beam was incident perpendicular to the DM surface. This feature aids in predicting the mirror shape to provide defocus and compensation for spherical aberration since circularly symmetric deformations are necessary. The normal incidence also allows larger spot sizes to be used when compared to off-axis beams. Software was developed in-house for controlling the DM and to supply feedback to enable optimization of the image acquired by the MRC 600 scan head.

The initial feedback used to optimize the shape of the DM was derived from the magnitude of the two-photon induced fluorescence signal. A region of interest on the image was selected and the brightness optimized to provide a psf which was, on average, the most suitable for that region thereby yielding the brightest, sharpest, image. By performing the optimization on the image properties, scope for using contrast or sharpness as merit functions, in addition to the fluorescent intensity, is generated. However, this does limit the optimization speed since for each change of shape of the mirror, a scan is needed to provide the feedback parameters. Several approaches to performing the optimization were considered, each being selected as appropriate to the sample. For samples which were nonbiological in nature and not prone to photo-bleaching with the laser powers being used, a hill climbing algorithm was applied. This could be started from a number of random actuator voltage values in an attempt to converge at the global, rather than a local, optimized mirror shape. In order to make the process quicker, groups of actuator voltages were altered simultaneously at first and in the later stages single actuator values were changed. The time for the entire process depended on the degree of optimization required and could vary between 30 seconds and 15 minutes. Clearly, as the changes in the actuator voltages grow smaller as the mirror shape approaches the optimal, there is a diminishing return in terms of image quality improvement. All optimizations were halted when the noise in the system became similar in value to the changes in the signal being produced by the increasingly small perturbations of the DM. With fairly inert samples, time could be taken to ensure optimization was complete and that the process did not assume prior information about the aberrations imparted by the optical train or the sample. However, when imaging biological samples, the minimum number of scans often need to be used in order to reduce undesirable effects attributable to laser scanning. In this case, a trade-off exists between the ultimate resolution achievable and the number of scans used in the optimization process. For many such samples, spherical aberration due to a mismatch in refractive index between the immersion fluid of the lens and the sample dominates. Sets of actuator voltages which provided aberration correction for various depths of index mismatched fluids were easily produced experimentally by performing optimizations on manufactured samples. Generally, these consisted of a cover slip and a fluorescent layer in which the optimization was performed with the fluid volume in between. The actuator voltage sets could be stepped through quickly in practice when imaging biological samples to determine the best set thus yielding significant improvements in only a few scans. Clearly this is advantageous not only in the respect that few scans have to be used but also that each actuator set has been thoroughly optimized to take into account an amount of sample induced spherical aberration in addition to any aberrations introduced by the optical train. A future objective is to link actuator sets to the z-drive movement so that spherical aberration can be automatically compensated for as depth into the sample is varied, this being done without any user intervention. Naturally, this needs to be done very carefully since deforming the mirror while compiling a z-section could otherwise distort the image.

Lateral scan aberrations can become significant in microscopy, these generally increasing with the scan angle. With the adaptive optics, it is possible to perform correction on a point by point basis. This scheme may be implemented in the future if much larger area xy slices are required. However, slower scan rates would be expected. If a scanning stage can be used, which is not suitable in many applications, then it is easier to use this in conjunction with the adaptive optics providing axial aberration correction through the objective in order to acquire large area images. This facility is also being built in to the current microscope for imaging other types of samples.

3. Experimental results

Investigations into the performance enhancements available using the adaptive optics were initially undertaken using artificial samples. Sub-resolution beads at varying water depths were scanned in the xy and xz-axis both with and without adaptive-optic correction. Figures 2(a) and 2(b) show xy slices at the center of a nominally 105nm diameter bead at a depth of 25.7µm when using the same laser power in order to highlight changes in the measured signal.

 

Fig. 2. xy scan of 105nm bead before (a) and after correction (b). Image dimensions are 5.0 µm × 2.3 µm.

Download Full Size | PDF

A xz section through a 105nm bead in water positioned just under the surface of the coverslip is shown in Fig. 3(a). In turn, Figs. 3(b) and 3(c) show xz sections through a 105nm bead at a water depth of 25.7µm without and with correction applied.

 

Fig. 3. xz scan of a 105nm bead just under the coverslip (a) and a bead at a water depth of 25.7µm before (b) and after correction (c). Image widths are 3.8µm with the scan depths being 2.1, 4.3 and 2.4µm, respectively.

Download Full Size | PDF

The effect of the reduction in the axial spread of the psf is clearly visible when comparing Figs. 3(b) and 3(c). The adaptive optic reduces the elongated psf caused by the refractive-index mismatch so that the deep bead appears similar to that attached to the underside of the cover slip, shown in Fig. 3(a). However, in order to obtain a better approximation to the axial psf, fluorescent plastic blocks separated from the cover slip by varying depths of water were used. The psf was then extracted by taking the derivative of the average integrated signal of each xy slice as the focal spot was scanned through the plastic-water interface. The results are shown in Fig. 4. In this graph, the FWHM of the axial psf is plotted as a function of the water depth. It can be seen that the adaptive optic initially helps to correct for aberrations in the optical system and then additionally for sample induced aberrations with increasing depth. A much narrower psf was achieved than with no correction, the increase with depth being due to the mirror running out of dynamic range. This problem has been confirmed when modeling the system using Zeemax and could be overcome by upgrading the current amplifiers driving the DM in order to obtain a larger maximum mirror deflection. Nevertheless, the adaptive optic extends the imaging depth attainable from approximately 3.4 to 46.2µm at a resolution defined by a FWHM axial psf of 1.25µm.

 

Fig. 4. FWHM of the axial psf as a function of water depth with and without correction.

Download Full Size | PDF

Having demonstrated considerable benefits when using artificial samples, the system was used to image a biological specimen, in this case smooth muscle from guinea pig bladder. To demonstrate the effectiveness of the adaptive optic, a ×40, 1.3 NA, oil immersion lens was used to image the specimen. Optimization of the deformable mirror was performed at an objective movement of 30.8µm into the sample. Figures 5(a) and 5(b) show a feature at this depth with the DM optimized for the surface (shape A) and at this depth respectively (shape B).

 

Fig. 5. Feature in tissue imaged using mirror shape A (a) and mirror shape B (b). Image dimensions are 32 µm × 32 µm.

Download Full Size | PDF

The peak signal of these Kalman 8 averaged slices was increased from a value of 168 to 235, a 40% gain. The 32×32µm image in Fig. 5(b) can also be seen to be sharper and in possession of more detail. Optimization of the adaptive optic mirror shape through the use of feedback from the fluorescence emission performed satisfactorily in uniform samples. This technique was carried over for use in the biological samples so its robustness as an optimization criterion was examined. Figure 6 shows the measured fluorescence as a function of depth in the sample when averaged over the whole area when using both mirror shapes A and B. Also included is the percentage improvement in the signal attained by using mirror shape B when compared to A. The general trends are what would be expected, as will be discussed.

 

Fig. 6. Signal vs. objective movement into smooth muscle cells.

Download Full Size | PDF

Examining the percentage improvement curve in Fig. 6, the initial “advantage” of shape B as opposed to shape A at the surface of the sample is false, this being due to the noptimized psf at this point. In this case, the signal intensity increases before that of the narrower psf provided by shape A. The steeper slope of the curve generated by mirror shape A as compared to that produced by shape B as the focussed spot entered the sample volume supports this observation. This demonstrates that in certain circumstances a larger psf could be created by the automatic ptimization function since it brings fluorescent features into the focal volume. It could present problems in samples with a high proportion of dark space where a different feedback for optimization would be required, such as sharpness or even contrast. This problem has not been observed in practice to date, however, since generally features are selected in the sample volume on which to perform the optimization. It is vident from Fig. 6 that mirror shape A clearly provides a better psf at shallower depths into the sample while shape B is better for those deeper in, the cross-over happening at around 20µm. The trend of the percentage improvement curve is a function of the biological sample as well as the psf. The dip at around 37µm corresponds to an area having a lower density of stained tissue which obviously diminishes the difference between the two signal curves.

4. Conclusions

A scanning galvo multiphoton microscope incorporating a deformable mirror and automatic optimization of the psf using intensity feedback from the two-photon induced fluorescence has been presented. The short path length and low number of intermediate optics between the sample and detector enables high photon collection efficiencies. Reduction in the axial psf has been demonstrated when focussing into samples, both artificially produced and biological, these having a different refractive index as compared to the immersion oil of the objective lens. The resolution of the microscope as defined by a 1.25µm FWHM of the axial psf was extended from approximately 3.4 to 46.2µm when focussing an oil immersion lens into a water medium. The dynamic range of the adaptive optic, soon to be upgraded, forms one of the limiting factors in the resolution enhancement with depth.