Two-photon fluorescence microscopy provides a powerful tool for deep tissue imaging.
However, optical aberrations from illumination beam path limit imaging depth and
resolution. Adaptive Optics (AO) is found to be useful to compensate for optical
aberrations and improve image resolution and contrast from two-photon excitation. We
have developed an AO system relying on a MEMS Deformable Mirror (DM) to
compensate the optical aberrations in a two-photon scanning laser fluorescence
microscope. The AO system utilized a Zernike polynomial based stochastic parallel
gradient descent (SPGD) algorithm to optimize the DM shape for wavefront correction.
The developed microscope is applied for subsurface imaging of mouse bone marrow. It
was demonstrated that AO allows 80% increase in fluorescence signal intensity from
bone cavities 145um below the surface. The AO-enhanced microscope provides cellular
level images of mouse bone marrow at depths exceeding those achievable without AO.
In a recent effort, researchers from Wellman Center of Photomedicine use fluorescence
signal provided by single- or two-photon excitation, second harmonic generation and
coherent anti-Stokes Raman spectroscopy (CARS) to illustrate the cell level detail of
mouse bone marrow . However, the several non-linear imaging techniques suffered on
a common base: signal degradation with deeper light penetration. The fluorescence signal
weakening from the mouse skull is caused by the decreased excitation light intensity.
With deeper imaging depth, the excitation light suffers tissue scattering, absorption and
optical aberration. The last one of the causes spreads the light intensity away from its
diffraction limited focal spot. In consequence, less fluorescence light is produced in the
enlarged focal volume. In this paper, I will introduce Adaptive Optics (AO), a system for
real time optical aberration compensation, to improve the non-linear fluorescence signal
in the mouse bone marrow imaging. A parallel stochastic gradient decent algorithm based
on Zernike polynomial is employed to control the deformable mirror in real time
Non-linear imaging is widely used in biological imaging, primarily because of its ability to image
through tissue to depth of a few hundred micrometers. Because two photons need to be
absorbed to excite a fluorophore in this instrument, the probability of fluorescence emission of a
detectable photon scales with the intensity squared of the beam. As a result, aberrations in the
beam path that reduce the peak intensity of the focused, scanned laser spot have a significant
effect on the instrument performance. Methods for reducing those aberrations should allow higher
resolution and detection sensitivity, and deeper tissue imaging.
In this paper, I will describe a non-linear imaging microscope that has an adaptive optics (AO)
subsystem to compensate for beam path aberrations. The AO system relies on a 140 actuator
deformable mirror, controlled using a stochastic gradient descent algorithm with feedback from a
fluorescence sensor. The controlled instrument will be used for in vivo imaging of mouse skin,
lymph nodes, and skull bone marrow at depths up to 500 &mgr;m.
In studying retinal disease on a microscopic level, in vivo imaging has allowed researchers to track disease
progression in a single animal over time without sacrificing large numbers of animals for statistical studies.
Historically, a drawback of in vivo retinal imaging, when compared to ex vivo imaging, is decreased image
resolution due to aberrations present in the mouse eye.
Adaptive optics has successfully corrected phase aberrations introduced the eye in ophthalmic imaging in
humans. We are using adaptive optics to correct for aberrations introduced by the mouse eye in hopes of
achieving cellular resolution retinal images of mice in vivo. In addition to using a wavefront sensor to drive
the adaptive optic element, we explore the using image data to correct for wavefront aberrations introduced by
the mouse eye. Image data, in the form of the confocal detection pinhole intensity are used as the feedback
mechanism to control the MEMS deformable mirror in the adaptive optics system. Correction for wavefront
sensing and sensor-less adaptive optics systems are presented.
We use a simplified mechanical/electrostatic model to describe the coupling between mirror and actuator. A WYKO interferometer is used to characterize the electromechanical performance of a MEMS deformable mirror (Boston Micromachines, Inc). We measured the voltage vs. deflection curves for the sample actuator with and without energizing the local adjacent neighbor actuators. This characterization results generated a quadratic and a linear equations to predict required voltage for actuators under different deflection profiles. We incorporated the MEMS mirror into a simple adaptive optics (AO) testbed. The system includes a near infrared superluminescent diode, a MEMS deformable mirror (DM), and a Shack Hartmann wavefront sensor (SHWS). The real time measurements provided by the SHWS (wavefront slopes) were the input to an integral controller. The controller was calibrated in situ by the typical method of determining an influence matrix which poke each actuator separately and measuring the resulting wavefront slopes at each lenslet. The control software then use the error signal between the current SHS positions and the desired positions, applied the characteristic model of the mirror, and determined the appropriate voltage to apply to each actuator, given the desired deflection for the surrounding actuators. The system was able to provide real time aberration compensation at loop gains of 0.3. A set of Zernike polynomial shapes were produced by DM under different loop gains to test the ability of control. A large proportion of the final wavefront shape could be achieved in a single iteration with a loop gain 1.0.
An important consideration in the design of an adaptive optics controller is the range of physical shapes required by the DM to compensate the existing aberrations. Conversely, if the range of surface shapes achievable with a DM is known, its suitability for a particular AO application can be determined. In this paper, we characterize one MEMS DM that was recently developed for vision science applications. The device has 140 actuators supporting a continuous face sheet deformable mirror having 4mm square aperture. The total range of actuation is about 4μm, achieved using electrostatic actuation in an architecture that has been described previously. We incorporated the MEMS mirror into an adaptive optics (AO) testbed to measure its capacity to transform an initially planar wavefront into a wavefront having one of thirty-six orthogonal shapes corresponding to the first seven orders of Zernike polynomials. The testbed included a superluminescent diode source emitting light with a wavelength 630nm, a MEMS DM, and a Shack Hartmann wavefront sensor (SHWS). The DM was positioned in a plane conjugate to the SHWS lenslets, using a pair of relay lenses. Wavefront slope measurements provided by the SHWS were used in an integral controller to regulate DM shape. The control software used the difference between the the wavefront measured by the SHWS and the desired (reference) wavefront as feedback for the DM. The DM is able to produce all 36 terms with a wavefront height root mean square (RMS) from 1.35μm for the lower order Zernike shapes to 0.2μm for the 7th order.