1.

Introduction

In vivo visualizing bone marrow has been a major attraction for biological research, especially stem cell research,^1,2 because studying biological systems as they evolve in their natural and physiological state could provide more relevant information than that in an in vitro environment. However, it is still big challenge for confocal fluorescence microscopy, which is widely used in biological research, to achieve optimal and diffraction-limited resolution in vivo imaging, especially in deep tissue.

Because biological samples are intrinsically turbid mediums (i.e., proteins, nuclear acids, and lipids) with optical properties characterized by strong multiple scattering and heterogeneity in its refractive index, and refractive mismatching between immersion media and biological samples³ they induce optical aberrations resulting in an enlarged focal spot within the sample and a concomitant deterioration of signal and resolution.^4,5 Therefore, the resolution and contrast of confocal microscopy is compromised in vivo. To observe clear bone marrow structure in vivo, correction of aberrations induced by heterogeneity in refractive indexes is necessary.

The aberrations could be compensated by adaptive optics (AO) technique, which has been widely used in astronomy and vision science. The typical AO systems employ a sensor to measure wavefront aberration in the image-path of the optical system and correct the aberration in a feedback loop. However, these methods require a point-like reference source, such as the guide star used in astronomical systems. In vision science, AO systems found their guide-star from the stable reflection of the retina, which is a somewhat translucent tissue. Currently, AO technology has been combined with confocal scanning laser ophthalmoscope⁶ and optical coherence tomography⁷ to image living human retina at the single cell level.⁵ They have been the basic tools in ophthalmology research.

However, in the field of microscopy, the situation is more complicated since the three-dimensional (3-D) structure of the specimen means that the reference source is generally far from point-like.⁸ In fact, the wavefront sensor would receive a multitude of wavefronts emitting from different parts of the specimen. Currently, several groups have conducted research on wavefront correction related to single-photon or multiphoton scanning microscopy^8–12 with different AO correction methods. Fluorescent protein or fluorescent beads have been used as the guide star to measure aberration of tissue. Live imaging of Drosophila embryos was also introduced by Tao et al.¹⁰ However, it is not always feasible to find appropriate fluorescent protein in different samples at different depths. Therefore, most of the adaptive microscope systems so far have used indirect methods for aberration measurement.⁸ Wavefront sensorless AO systems operate by sequentially modulating the AO corrector and maximizing a feedback signal according to particular optimization algorithms. The modal wavefront sensor approach proposed by Booth et al.¹³ sequentially measured an image quality metric with a group of setting trial aberrations. The method was investigated in detail in Ref. 14 and works well with small aberrations.¹¹ Model-free search methods contain stochastic, local, or global search methods,¹⁵ such as genetic algorithm, hill-climbing algorithm, and simulated annealing algorithm. Other approaches used a wavefront measurement based on sequential measurements of the wavefront error in each segment of the aperture.⁹

Due to the flow characteristics of blood in vivo in the bone marrow vasculature, guide star methods are not suitable because it is not easy to fix the fluorescent guide star in vessels. Therefore, to achieve a stable and accurate wavefront correction for mouse bone marrow in vivo, we chose a sensorless AO scheme and introduced stochastic parallel gradient descent (SPGD),¹⁶ which has been used in two-photon microscopy, as the wavefront optimization algorithm.¹⁷ The SPGD proposed by Vorontsov has been used in other areas of AO and has been verified to be the fastest search method. Here, we demonstrate an AO confocal fluorescence microscope with SPGD optimization algorithm. A deformable mirror (DM) was used as wavefront corrector to compensate system- or specimen-included aberration. In vivo imaging experiment of mouse bone marrow vasculature was accomplished. The results have demonstrated that the aberration correction with the SPGD algorithm has improved the image quality and fluorescence signals.

2.

Materials and Method

2.1.

Experimental Setup

Figure 1 shows the experimental configuration used for AO aberration correction and in vivo fluorescence imaging. The beam light emitted from a 638-nm diode (LD) with output power $\sim 30\mathrm{mW}$ was collimated. The expanded beam was imaged onto a DM with 37 actuators. Then, two scanning mirrors were used to scan the laser in the $X$ and $Y$ directions (the objective axis being the $Z$ direction). To keep the phase correction stationary during beam scanning, the DM was conjugated to the $X$ scanner and the $X$ and $Y$ scanners were conjugated to each other, which was then conjugated to the back pupil plane of the objective. All the optical conjugation was achieved by custom-designed spherical mirrors. The fluorescence emission was collected by an objective lens (Olympus UA, water immersion, $20\times$, $\mathrm{NA}=0.5$, Japan). The working distance of the objective is 3.5 mm. A band-pass filter (Semrock, $\mathrm{FF}01\text{-}670/30\mathrm{nm}$, Rochester, New York) was placed before the photomultiplier tubes (PMT, Hamamatsu, H7422-20, Japan) to filter illumination light. The pinhole used in this microscope was $50\mu \mathrm{m}$. Finally, the fluorescence from the specimen was detected by PMT. Specimen scanning in the axial $Z$ direction was enabled by a motorized translation stage. The theoretical lateral resolution and axial resolution are 0.55 and $3.8\mu \mathrm{m}$, respectively. A Hartmann wavefront sensor was used to measure system aberration and for calibration of the Zernike basis mode.¹³ The imaging speed of the microscope was 30 Hz.

Fig. 1

Schematic diagram of the adaptive optical confocal fluorescence microscopy.

3.

Results

First, a Hartmann wavefront sensor was used to measure system aberration using a fluorescence bead attached to the glass slide as the guide star and the system aberration was compensated before imaging experiment in vivo. Root mean square (RMS) of system aberration was $0.142\mu \mathrm{m}$. After aberration correction, RMS went down to $0.079\mu \mathrm{m}$. The corresponding point spread functions of microscopy before and after AO correction are also shown in Fig. 2.

Fig. 2

(a) Wavefront of system aberration measured by Hartmann sensor. (b) Point spread function (PSF) of system before system-induced aberration correction. (c) Wavefront of residual aberration measured after correction of system aberration. (d) PSF of system after aberration correction.

Using molecular probes injected into mouse vein, fluorescence images of a bone marrow vasculature $38\mu \mathrm{m}$ below the surface of the skull was obtained, and aberration correction was accomplished as shown in Video 1.

Video 1

Video of the bone marrow vasculature at $38\text{-}\mu \mathrm{m}$ depth during aberration correction (MOV, 7.31 MB) [URL: http://dx.doi.org/10.1117/1.JBO.19.8.086009.1].

Lateral images of the vasculature before and after AO correction are shown in Fig. 3(a). As seen from this figure, in addition to a great enhancement of the fluorescence signal, the details of the vessel are more abundant, since the aberration correction could improve resolution of the microscope. A sharper image of one cell in the vasculature, shown in magnified image of Fig. 3(a), was achieved.

Fig. 3

(a) Lateral fluorescence bone marrow vasculature images $38\mu \mathrm{m}$ below the skull surface before and after aberration correction. (b) Fluorescence signal profiles along the black dot line in (a). (c) Signal profiles along the white line in (a). (d) The final corrective wavefront of deformable mirror (DM). (e) Metric value of images during aberration correction process.

The signal profile along the white line in Fig. 2(a), shown in Fig. 3(c), reveals that the improved signal and resolution result in substantially enhanced contrast. The metric values of images during the aberration correction process are shown in Fig. 3(e). After 150 iterations, the metric value of images increased threefold with the same excitation power. It takes about 5 s to complete the correction for our microscope. The compensation profile of the DM is shown in Fig. 3(d).

It is a common observation that image resolution and contrast degrade with imaging depth. In Fig. 4, we summarize the fluorescence images of the vasculature at depths down to $75\mu \mathrm{m}$ without AO correction and with AO correction. The video during aberration correction is shown in Video 2. We also present the signal profile along the dotted line in Fig. 4(a), which reveals the improved signal and resolution results with enhanced contrast, therefore, more vasculature detail can be defined after correction.

Fig. 4

(a) Lateral fluorescence bone marrow vasculature image $75\mu \mathrm{m}$ below the skull surface before and after adaptive optics correction. (b) Fluorescence signal profiles along the dot line in Fig. 2(a). (c) The final corrective wavefront of DM. (d) Metric value of images during aberration correction process.

Video 2

Video of the bone marrow vasculature at $75\text{-}\mu \mathrm{m}$ depth during aberration correction (MOV, 3.50 MB) [URL: http://dx.doi.org/10.1117/1.JBO.19.8.086009.2].

The metric value of the image after AO correction, shown in Fig. 4(d), increased nearly twofold with the same excitation power. It takes about 150 to 200 iterations to complete the correction. It also reveals that the closed-loop stability decreases, because the index-mismatch aberration increases with image depth and perturbations during SPGD iterative process may go beyond the capability of the DM to compensate. The compensation profile of the DM is shown in Fig. 4(c).

The aberration corrected images obtained with the microscope show enhanced contrast and improvement of fluorescence intensity. The AO could be widely beneficial in confocal microscopes that often suffer from the detrimental effects of system- and specimen-induced aberrations, especially for in vivo imaging.

4.

Discussion and Conclusion

In vivo bone marrow vasculature imaging has played an important role in research regarding the dynamic interactions between tumor/stem cells and the bone marrow microenvironment. We have demonstrated confocal fluorescence microscopy with a sensorless AO system which could be used for in vivo imaging of mouse bone marrow. The results show improved imaging depth and imaging resolution. With aberration correction, finer bone marrow cavity structure information and more accurate locations of cells could be observed.

The custom built confocal fluorescence microscope could perform real-time imaging at the video level, which is ideally suited for in vivo imaging and long-term observation of vascular cells within the bone marrow.

Because of the flow characteristics of blood, guide star methods are not suitable because it is not convenient to fix injected fluorescent beads or find fluorescent proteins in vessels in vivo. Therefore, a wavefront sensorless scheme was chosen in the microscopy. Compared to conventional AO using direct wavefront sensing which needs a guide star and wavefront sensor, such as the Hartmann sensor, to measure the aberration of the imaging-path from the guide star, sensorless AO has a simpler system architecture and can non-invasively correct aberrations for different depths of biological tissue, which is very important in the field of microscopy.

The aberration magnitudes of bone marrow during in vivo imagine have significant variations at different imaging depths. Therefore, the AO method should be adaptive to correct different magnitudes of aberrations. An optimization algorithm based on a random search was selected because of its advantages of better convergence robustness for aberrations of various magnitudes. Compared to the modal sensor algorithm, it is simple and does not require a system calibration process.²¹ The modal sensor needs a different control basis for different magnitudes’ aberrations^22,23 and also requires a complex system calibration process.²¹ Correction speed also has important significance for fluorescence imaging because of the photo bleaching issue for fluorescent dyes. Considering the range of the algorithm application and correction speed, we choose the SPGD algorithm, which has robust convergence characteristics for different aberrations and can complete the aberration correction in seconds in a confocal fluorescence microscope, which is first time this has done to the best of our knowledge.

The experiment of in vivo mouse bone marrow imaging was completed. Fluorescence images of blood vessels at different depths are obtained and image contrast has been significantly improved after correction. For the blood vessel at the depth of $38\mu \mathrm{m}$, the convergence process was relatively stable and more blood cell morphology can clearly be observed after correction. For the blood vessel at the depth of $75\mu \mathrm{m}$, enhancement of the fluorescence intensity after correction was also observed, but the stability of the convergence process decreased. The main reason is that the fitting ability of the DM is limited by the drive stroke volume; therefore, some drives were saturated with deeper imaging position. In addition, the lower resolution of the image of a deeper blood vessel after correction was observed, mainly due to more severe scattering.²⁴ The results prove that the image-based wavefront sensorless AO with the SPGD algorithm could meet the requirements of aberration correction for in vivo imaging in real-time confocal microscopy.

For the optimization of the first 15 Zernike modes, typically 100 to 200 images are necessary to achieve good correction results. It takes 5 to 7 s to accomplish aberration correction, which is not conducive in the cases with more stringent time requirements. The convergence time is mainly limited by the imaging speed of the microscope and the convergence rate of the SPGD algorithm. An improved SPGD algorithm or combining it with other optimization algorithms such as the modal algorithm may be a potential solution to accelerate the process of aberration correction. This will be the next focus of our future work.

To conclude, by incorporating an SPGD-based AO in a confocal fluorescence microscope, we demonstrated that the correct use of AO reduces specimen-induced aberrations in vivo in the mouse skull marrow. The fluorescence images with aberration correction show enhanced contrast and improvement of fluorescence intensity. Therefore, AO would play an important role in the application of many physiological imaging techniques in vivo. While the confocal fluorescence microscopy could perform real-time imaging of blood vessels. Not only can the 3-D structure be observed, but the migration of blood cells or other observations of immune cells in the blood vessels can also be observed. It would be the ideal tool for study of vascular dynamics in future.