Smart illumination for 3D-imaging of biological tissues FOCUS
Figure 4. Smart-scanning. Sometimes, the signal of interest occupies a small fraction of the sample volume. Such is the case for example when cells are organized as a curve surface( left). We can then estimate a spatial model of the structure( computed surface in red), and concentrate illumination to a thin shell around this structure of interest( right).
contribution of the out-of-focus regions varies much less. As a result, the variance processing washes out the out-of-focus background.
In both the confocal and RIM procedures, the out-of-focus contributions are discarded physically or numerically: we end up using only the photons thought to originate from sources in the focal plane. This approach is justified if only one slice of the sample is imaged. On the other hand, if one wants to record a volume image of the sample, discarding the photons arriving from fluorophores above and below the focal plane is not a good idea as they carry information on the three-dimensional( 3D) structure of the sample. This led us to develop a 3D version of RIM that uses the 3D nature of speckle and 3D data processing to exploit the information carried by these out-offocus photons( see Fig. 3).
In 3D-RIM, we sequentially image the volume of the sample while keeping the 3D speckle pattern fixed. Instead of moving the sample, axial scanning is carried out by a remote-focusing unit placed in the detection arm, such that the illumination stays registered and acts as a stable 3D probe. The structuring of the speckle along optical axis can then be used to improve sectioning. See the illustration in Fig. 3: when imaging a given plane, the fluorophores above or below this planes provide out-of-focus photons appearing as smears on the image. If we have knowledge of the 3D stack, however, we can attribute those out-of-focus photons to their source— effectively
reassigning them to the proper plane. Thus there is no need to suppress the smears: we use them and put the photons back where they belong. Mathematically, this reassignment is performed through a deconvolution step before computing the 3D variance.
The consequences are that 3D-RIM keeps RIM’ s robustness in thick, aberrated tissues while adding stronger axial sectioning and better photon budget— fewer photons are thrown away by the variance step. In practice, the same level of detail can be obtained with lower illumination power, making the method gentler for live samples.
EXTENDED DEPTH OF FIELD While faster than the confocal, structured illumination microscopies, including RIM, are slower than the widefield microscope as they require the acquisition of multiples images of the sample under different illuminations. In addition, thick 3D samples require many sequential slices to reconstruct a volume. Each newly acquired image also comes with a camera " dead time " linked to reading out of the information on the sensor. When biology is fast and the scientific question tolerates a projective view along z, we reasoned that compressing the whole volume into a single exposure becomes a powerful alternative. This can be done through extended depth imaging [ 6 ].
During each camera exposure, the focal plane is swept rapidly through the specimen using a tunable element( for example, an electrically tunable lens). The sensor therefore integrates fluorescence from the full axial range, yielding one extended-depth frame. Repeating this with a set of different speckle illuminations provides the data for variance-based reconstruction, which removes the characteristic blurred background that normally mars extended depth images. We could thus retrieve super-resolved details in the projection.
Compared with acquiring a full 3D stack with standard RIM, EDF-RIM reduces the number of readouts and can be about ten times faster, while maintaining lateral resolution. The price to pay is that axial information is compressed into
Photoniques 134 I www. photoniques. com 39