Conceived and designed the experiments: FB CB FM. Performed the experiments: PP CDC. Analyzed the data: FB CB. Contributed reagents/materials/analysis tools: CB. Wrote the paper: FB FM. Designed the software used in PSF minimization: CDC.
The authors have declared that no competing interests exist.
Graded Index (GRIN) rod microlenses are increasingly employed in the assembly of optical probes for microendoscopy applications. Confocal, two–photon and optical coherence tomography (OCT) based on GRIN optical probes permit in–vivo imaging with penetration depths into tissue up to the centimeter range. However, insertion of the probe can be complicated by the need of several alignment and focusing mechanisms along the optical path. Furthermore, resolution values are generally not limited by diffraction, but rather by optical aberrations within the endoscope probe and feeding optics. Here we describe a multiphoton confocal fluorescence imaging system equipped with a compact objective that incorporates a GRIN probe and requires no adjustment mechanisms. We minimized the effects of aberrations with optical compensation provided by a low–order electrostatic membrane mirror (EMM) inserted in the optical path of the confocal architecture, resulting in greatly enhanced image quality.
Optical microscopy for
Applications for in–vivo two–photon fluorescence microscopy using scanned GRIN probes have been used for basic research
In our prior work we described a compact infinity–corrected GRIN objective, with 0.5 numerical aperture (NA) in water, suitable for microendoscopy (0.5 mm diameter), which we assembled from commercially available components
an AO module based on an EMM
a GRIN fiber objective mounted on the microscope standard objective receptacle
a calibration system based on an imaging camera illuminated by an optical relay reproducing a magnified view of the field covered by the GRIN objective
A standard two–photon fluorescence microscope was modified by insertion of an AO module, a PSF calibration system and a GRIN fiber objective (grey boxes).
An important constraint in this AO application was to maintain the same adaptive mirror correction action, in terms of impulse response function, over the whole view field explored by the microscope scanning head, avoiding light vignetting. In principle, one could recreate a suitable field invariant pupil after the confocal scanning head using custom designed optics. Unfortunately this proved impossible in our configuration, due to constraints imposed by the underlying commercial architecture. The simplest solution, and the one we adopted, was to intercept the laser beam before it entered the scanning head (
Detailed diagram of the AO module interposed on the input laser beam before the microscope scanning head.
A view of the mounted objective with its main specifications is presented in
Top: the objective was assembles using a commercial aspheric lens coupled to the GRIN fiber and embodied in an anodized aluminum mount with a standard microscope collar. Bottom: Zemax spot diagrams for the Deformable Mirror plus GRIN objective combination in water; left box, the nominal on–axis Airy disk is about 2 µm; center box: beam distortion at 80 µm from center; right box: simulated correction performed by the EMM.
The calibration system was assembled using commercial relay optics and inserted on the fluorescence beam returned by the GRIN objective before the photomultipliers that formed the so called
In a traditional AO application, a classical example is the astronomical case
The first operation is detailed in
After construction, the
and with index
First eight flexure modes (top) and corresponding voltage patterns (bottom) used for PSF optimization. Modes were generated by sequentially pulsing each electrode at minimum voltage and recording the corresponding wavefront phase with the Shack–Hartmann camera.
The second step, the iterative optimization process of the point spread function (PSF), required the immersion of the GRIN objective in a fluorescent solution to retrieve an image of the PSF on the calibration system camera. The key idea is to minimize PSF image size in a loop, whereby low order flexure modes weight factors are perturbed and the resulting PSF is estimated. Iterative minimization was based on a merit factor extracted from each image after a linear fit of a two–dimensional Gaussian to the fluorescence spot profile recorded by the calibration camera:
The optimization metric described by Eq. 4 was preferred because it involves both the intensity and shape of the sampled PSF. The Strehl ratio, the most obvious choice, was considered unsuitable because strictly dependent on PSF intensities, which were relatively unstable due to noise in the detected signal during long optimization trials. Compared to simple ‘spot size’ measurements, the merit factor ε proved to be less susceptible to noise and provided the most stable and fast convergence. In order to minimize the camera noise pattern error, ε was computed on running groups of ten consecutive PSF images. ε was than used by a C code implementation of the classical downhill simplex minimization method
Minimizations were tested both using the first eight EMM flexure modes, as well as using just the first four, with similar results; results of a typical PSF minimization trial with four modes are shown in
Behavior of merit factor ε (top) and of first four eigenmodes weight factors (bottom) versus minimization step. The final pattern commanded to the EMM after optimization is shown as inset in the top panel.
Top, AO module not inserted in the optical path; middle: module inserted with EMM at rest position; bottom: typical result after AO optimization.
Merit Factor ε | Mode 1 | Mode 2 | Mode 3 | Mode 4 | |
Average Value | 1,898 | 0,108 | −0,047 | −0,179 | −0,200 |
RMS | 0,011 | 0,065 | 0,022 | 0,069 | 0,041 |
No AO | AO at rest | AO | |
FWHM µm | 2.39 | 3.71 | 1.98 |
FWHM Gaussian fit µm | 3.46 | 5.09 | 2.88 |
Ellipticity | 0.91 | 1.14 | 0,98 |
As a final test of our imaging apparatus, we imaged fluorescent micro–beads (1.0 µm diameter, peak emission around 515 nm) excited by the two–photon laser tuned at 830 nm with and without AO assist. As shown in
Whole view field before (left) and after (right) AO correction with parameters retrieved during optimization. The field spans about 80×80 µm.
A chain of four 1 µm fluorescent spheres imaged before (left) and after (right) AO application.
Thanks to microendoscopy, the penetration depth of laser–scanning microscopy into tissue can be increased up to the centimeter range
In the current study, we coupled a compact GRIN fiber objective, described in our prior work
a calibration procedure and optical architecture to insert the AO components in the microscope optical path
a computer algorithm to optimize the performance of the system
Tests with calibrated samples highlighted:
compatibility with the results obtained during system optimization phase
coverage of the full field of view with a corrected and uniform PSF.
Two critical factors limited the performance of our system, both due to the necessity of interfacing the AO module with a commercial multiphoton architecture. Firstly, the mismatch between the diameter of the EMM and that of the incoming laser beam forced us to introduce a beam expander and a corresponding beam compressor in the light path preceding the scanning head of the microscope. These extra optical components reduced the power of the laser beam reaching the sample and significantly complicated the alignment of the system. These problems can be solved using MEMS based mirrors with smaller and more numerous active elements, which would not require modification of the laser beam. The second critical point was the poor photon–capture capability of the commercial two–photon microscope, which was primarily due to the native remote positioning of the photodetectors, far away from the back focal plane of the objective. Also this limitation could be easily circumvented by a suitable redesign of the optical paths, as well as by adopting state–of–the art photosensors. Despite these shortcomings, the tests we performed with fluorescent microspheres clearly indicate that AO can substantially improve image quality by minimizing the effects of intrinsic and extrinsic low order aberrations in fluorescence multiphoton microendoscopy.
FluoSpheres® carboxylate–modified microspheres, 1.0 µm, yellow–green fluorescent, with peak emission around 515 nm (Cat. N. F–8823, Invitrogen) were imaged by the system described above a retro–fitted to Biorad Radiance 2100 confocal microscope mounted on a Nikon Eclipse 600 upright fluorescence microscope and fed by a mode–locked Ti:Sapphire laser (Tsunami, Spectra–Physics/Newport Corporation, Irvine, CA, USA).
The naked GRIN fiber was purchased from GRINTECH Gmbh. It was based on a ¼ pitch device with a 0.5 in–water NA coupled to a ¾ pitch relay lens with a 0.2 NA entrance. The overall probe (0.5 mm diameter) was inserted and epoxy glued in an steel capillary tube (0.7 mm outer diameter and 10 mm length) for light shielding and mechanical protection.
The AO module was based on an EMM with 37 electrodes controlled by high voltage amplifiers and interfaced to a PC using a 12–bit USB controller (Flexible Optical B.V., Rijswijk ZH, The Netherlands).
(TIF)
Images in