A detailed description of our OPT instrument depicted in
can be found elsewhere [24
]. In short, the specimen is placed in a capillary (C, BRAND, Wertheim, Germany, Blaubrand-intraMARK), and put in a custom-built bath (B) filled with glycerol. The walls of the bath are made of borosilicate microscopy cover slips matching the refractive index of both capillary and glycerol. For fluorescence excitation, the light of a super bright blue LED (LED1, Philips Lumileds Lighting, San Jose, CA, USA, Luxeon LXHL-LB3C 3W Star, typical luminous flux 23 lm) is focused by a lens (L, Carclo Technical Plastics, Slough, Berkshire, U.K., 4° beam divergence), and the excitation spectrum is narrowed by a filter (F1, Semrock, Rochester NY, USA, FF01-472/30). Fluorescence light is detected by a long working distance objective lens (OL, Mitutoyo Corp., Kanagawa, Japan,), the numerical aperture NA of which can be adjusted by an iris diaphragm (I) to typically NA = 0.2. Residual excitation light is blocked by the emission filter (F2, Semrock BrightLine HC 531/40 for GFP, Chroma ET 605/70 for dsRed), and the signal is focused onto a CCD camera (Andor Technology, Belfast, Northern Ireland, Ixon DV885) via the tube lens (TL, Infinity, Boulder CO, USA, InfiniTube FM). Alternatively, transmitted light may be recorded using a super bright white light LED (LED2, Philips Lumileds Lighting, Luxeon LXHL-LW6C 5W Star, typical luminous flux 120 lm). Its light passes through a diffusor (D) before illuminating the sample.
Fig. 1 OPT setup: The capillary (C) with the specimen is placed in a bath (B). White light (LED2) passes a diffusor (D) before illuminating the sample. Fluorescence excitation is achieved by blue light (LED1) which is focused (L) and filtered (F1). The light (more ...)
In both illumination modes we acquired 500 projection images over 360°. Rotation of the capillary around its center (z-
)axis with an angular step size of θ
= 0.72° was achieved by a rotation stage (8MR180, Standa, Vilnius, Lithuania). As specimens, we used Caenorhabditis elegans
) and Parhyale hawaiensis
). The C. elegans
transgenic zdIs5, expressing mec-4::GFP in touch neurons, was maintained as described in [27
]. Prior to the experiment, the worm was anesthetized for 15 min in 20 mM sodium azide, and then embedded in halo carbon oil to obtain a proper refractive index match to the capillary. The worms used in these experiments had a length of around 1 mm, and a mid-length diameter between 30 and 60 µm. Imaging was performed with a 10x objective lens (working distance 33.5mm, focal length 20mm, depth of field ca. 20 µm). With P. hawaiensis
, immobilization was precipitated using a low concentration of clove oil (0.04%) in artificial sea water. After 5 min, the specimen was transferred to the capillary filled with glycerol. Typically, these animals had a length from head to tail of roughly 2.5 mm and a maximum diameter (including extremities) of about 0.6 mm when inside the capillary. Imaging of P. hawaiensis
was performed using a 5x objective lens (working distance 34mm, focal length 40mm, depth of field ca. 150 µm). Since the depth of field was adjusted prior to each experiment, only typical values are given.
Although large efforts have been put in the design of the instrument and the specimens holder [24
], not all mechanical instabilities could be eliminated, resulting in a drift of the complete image (corresponding to about 1 µm per hour in the object space), even when the stages were stationary (i.e. no actuation). However, movement of the sample is also expected when imaging animals in-vivo
. Hence, post-acquisition correction methods need to be provided to address these issues and to correct for motion-induced blur in the final reconstructions. In general, all the correction methods described here assume the drifts to be constant for a given projection, i.e. each acquired 2D projection can be corrected by a 2D shift vector
) with independent components in y- and z-direction. However, we assume
to be constant for all pixels of that projection data, and only dependent on the rotation angle θ
. This approach can of course be generalized by dividing each projection in multiple patches, and by treating these patches individually, since similar patching has also been shown for high-resolution truncated insets in a low resolution full reconstruction [28
]. Fortunately, the movement is generally slow resulting in a strong correlation of the shifts between subsequent projections. All of the methods described below are implemented in the MATLAB programming environment (The MathWorks Inc., Natick, USA) and make use of the free DIPimage toolbox [29
]. Since the lateral and the longitudinal components of the shifts are independent, we can derive correction methods for each component separately as described below.