There is an increasing trend in biomedical research toward in situ studies, both ex vivo and in vivo, employing three-dimensional (3D) structural and functional imaging. For samples such as small animals, embryos, and engineered tissue in the “mesoscopic” ($\sim 1–10\mathrm{m}\mathrm{m}$) regime, a variety of imaging techniques have been developed, including optical projection tomography (OPT) [1], scanning laser optical tomography (SLOT) [2], selective plane illumination microscopy [3], and ultramicroscopy [4].

OPT, also known as optical computed tomography (CT) [5], is the optical equivalent of x-ray CT, in which the 3D structure of a rotating sample is reconstructed from a series of wide-field 2D projections acquired at different angles. Typically a filtered backprojection (FBP) algorithm is used for image reconstruction [6], although iterative techniques have also been developed [7]. A key advantage of OPT compared to x-ray CT for imaging biological tissue is the use of optical radiation with its rich spectroscopic contrast, including absorption, fluorescence intensity, and fluorescence lifetime [8]. Unfortunately, compared to x rays, the longer wavelength of optical radiation also leads to strong optical scattering and diffraction, which compromises the assumption of straight-ray projection in the FBP image reconstruction. To address the issue of scattering, it is necessary to apply OPT to samples that have been rendered transparent by chemical clearing [1,9] or that are inherently transparent [10,11]. To accommodate the constraints imposed by diffraction-limited imaging, OPT is typically configured such that the volume to be imaged is confined within the depth of field (DOF) of the imaging system, thereby approximating parallel ray projection. This constraint results in a trade-off between sample size and achievable resolution as the DOF scales inversely with the square of the numerical aperture (NA) of the imaging lens while the optical resolution improves proportionally with the NA. The potential application of OPT to intact mesoscopic samples for biomedical research, especially for in vivo studies, has stimulated interest in optimizing the image quality, resolution, and acquisition time of OPT [12–14].

In many “standard” OPT systems this trade-off is mitigated by arranging for the DOF to extend through half of the sample by locating the focal plane (FP) a quarter of the way into the sample, as depicted in Fig. 1(a), rather than at the axis of rotation. Thus the front half of the sample is imaged under the parallel ray approximation, which interrogates the entire sample during the course of a full rotation. To realize higher-resolution imaging over larger volumes, the DOF for a given NA can be effectively extended by axially scanning the FP through the sample [15]. While this can provide a significant resolution improvement for OPT of smaller samples, it is less attractive in the mesoscopic regime due to the larger focal scanning ($\sim \text{millimeter}$) distances required and the concomitant issues of scanning speed and stability.

 

Fig. 1. Schematic of (a) “standard” and (b) dual-axis OPT system setup (inset shows optimal DOFs and the FPs of two imaging systems). DOF, depth of field; EF, emission filter; AP, aperture; L1, 25 mm focal length lens; L2, 50 mm focal length lens; FP, focal plane; $\varphi$, sample diameter.

Download Full Size | PDF

To ameliorate this trade-off and image larger samples with increased NA OPT systems, we describe an angular multiplexing technique that decreases the required data acquisition times by increasing the light collection efficiency. Essentially we employ multiple imaging systems in parallel that can acquire OPT data sets simultaneously at different FPs within the sample. Here we report a dual optical imaging system acquiring OPT data at orthogonal angular projections. For a sample of a given size, this permits an increase in NA compared to the standard OPT configuration while maintaining the approximation to parallel ray projection if the DOF of each imaging arm extends throughout a quarter of the specimen with the FPs of the imaging arms being located at $1/8$ and $3/8$, respectively, along the radius of the sample, such that together they cover the front half of the sample. If the pixels of the imaging detector (e.g., CCD) sample the image sufficiently finely such that the image resolution of the system is limited only by the NA of the objective lens, then this dual-axis approach should permit a corresponding improvement in resolution by a factor of $\surd 2$. Furthermore, the light collection efficiency is improved by the increased NA and the use of two imaging arms, increasing collection efficiency by a factor of 4 compared to “standard” OPT. In practice, however, the range of pixel number and density for image sensors is limited and the image resolution and DOF do not depend only on the NA and imaging wavelength but are related by Eq. (1) [13,14]:

Here we demonstrate angularly multiplexed OPT using the experimental configuration depicted in Fig. 1(b). In brief, the sample was mounted under a rotation stage (T-NM17A200, Zaber Technologies, Inc.) and suspended in a refractive-index matched environment. A commercial 473 nm laser (Cobolt Blues™) was used to provide wide-field excitation. The sample was imaged onto two CCD cameras (Clara, Andor Technology plc, $1040\times 1392$, 6.45 μm pixel size, cooled to $-20°\mathrm{C}$) using two orthogonal imaging systems (L1, 25 mm achromatic doublet lens; L2, 50 mm achromatic doublet lens and an adjustable aperture) via appropriate EFs. The effective NA was adjusted using the apertures (AP) in the back FPs of lenses L1. The maximum field of view (FOV) of the system was $3.35\mathrm{m}\mathrm{m}\times 4.49\mathrm{m}\mathrm{m}$.

Figures 2 and 3 show data from a cylindrical phantom of 3 mm diameter comprising a low concentration suspension of fluorescent beads in 2% agarose, with an average bead diameter of 4.2 μm and excitation/emission maxima at 505 and 515 nm, respectively (F8859, Life Technologies Ltd.). This was suspended in a water-filled cuvette and imaged through a $520\pm 35\mathrm{nm}$ EF. Image data were acquired with an NA of 0.024 for the “standard” OPT system with a DOF of 1.4 mm and with an NA of 0.033 for each arm of the dual-axis system, which gave a DOF of 0.78 mm. Images were acquired by the CCDs at 0.9° angular intervals over a full rotation of the sample with 0.5 s integration time for both systems and reconstructed using an FBP algorithm. Figure 2 shows the resulting images of a bead located 0.43 mm from the rotation axis. For the “standard” single-axis OPT system, the FP was located 0.69 mm from the axis of rotation, while for the dual-axis OPT system the FPs were located at 0.37 and 1.05 mm, respectively, from the rotation axis. Since the beads are smaller than the optical resolution, these reconstructed images indicate the resolution of the OPT systems. The resolution improvement of the dual-axis OPT system is confirmed by the normalized line sections through the bead center shown in Fig. 3, with accompanying Gaussian fits providing full width half-maxima (FWHM) of 13.4 and 10.8 μm for standard and dual-axis OPT systems, respectively. Figures 2 and 3 also highlight the significant improvement in signal-to-background ratio that results from both the $\sim 3.8\times$ increase in light collection efficiency and the improved resolution.

 

Fig. 2. Reconstructed $X–Y$ and $X–Z$ image slices of a bead acquired with standard OPT (a), (b) and dual-axis OPT (c), (d).

Download Full Size | PDF

 

Fig. 3. Line plots and Gaussian fits through the $Y$ axis of the reconstructed bead in Fig. 2 for the dual-axis (D, red line) and standard (S, black lines) OPT systems with the latter showing relative (solid) and normalized (dashed) line intensity data.

Download Full Size | PDF

To further analyze the resolution improvement, a number of beads at different distances from the axis of rotation (0.17, 0.34, 0.43, 0.65, 0.98, and 1.39 mm) were analyzed and the measured average FWHM of these beads in the standard and the dual-axis OPT systems were $13.8\pm 0.8\mathrm{\mu m}$ and $11.1\pm 1.2\mathrm{\mu m}$, respectively, showing a $\sim 20\%$ improvement. The theoretical resolution (FWHM of Airy pattern) at best focus (i.e., limited by the NA of the objective lenses) was 11.1 and 8.1 μm for NAs of 0.024 and 0.033, respectively. The observed values are larger than the theoretical calculations because these OPT systems operate just within the sampling limit of the PSF and the reconstruction process involves interpolation. Nevertheless, we have demonstrated a significant improvement in image resolution and signal-to-background ratio that can be further improved using imaging detectors with more resolution elements and appropriate magnification.

To further illustrate the impact of the dual-axis OPTsystem, we performed imaging on the tail section of a fluorescent Casper:Fli-EGFP transgenic zebrafish imaged at 54 days postfertilization (dpf). Images were acquired at 200 angular projections with a CCD integration time of 0.1 s for both systems to maintain the total acquisition time at $\sim 2.5\min$ (made up of 20 s of photon detection and 131 s of stage movement, CCD setup, and readout). Figures 4(a) and 4(b) show autoscaled maximum intensity projections (MIPs) of the reconstruction from the dual-axis and the standard OPT systems, respectively, while Fig. 4(c) shows the same MIP for the standard OPT reconstruction displayed on the same intensity scale as Fig. 4(a). We note that, unlike zebrafish embryos, mature fish exhibit significant optical scattering. Nevertheless, we can obtain images near the surface of the fish that illustrate significant improvement in the signal level and contrast for the dual-axis system, as confirmed by the line sections shown in Fig. 4(d). We note this improved imaging efficiency could be used to reduce the acquisition time of the dual-axis OPT system to achieve the signal-to-noise ratio of the standard approach, but with a proportional reduction in illumination dose.

 

Fig. 4. Reconstructed (maximum intensity projection) images of the tail of a 54 dpf zebrafish acquired with (a) dual-axis and (b), (c) standard OPT plotted for comparison with normalized intensity (a), (b) and absolute intensity (a), (c) scales. (d) Intensity line profiles from (a), (b) as indicated, where the standard OPT intensity data have been multiplied by 2 for clarity.

Download Full Size | PDF

In conclusion, we have shown how angular multiplexing can improve the image resolution and collection efficiency of OPT by demonstrating a dual-axis OPT system acquiring image data at orthogonal angular projections that permits imaging with increased NA for a given sample. This approach can be used to image absorption as well as fluorescence and can be extended to more than two imaging axes, thereby further reducing image acquisition time and the corresponding light dose while improving spatial resolution and enhancing the ability to image larger samples. It could also be adapted for SLOT to improve the spatial resolution. We note that the orthogonal imaging systems could be configured to collect light from the same sample volume to simply increase the imaging speed, e.g., for intravital imaging.