Limits...
Optical Sectioning and High Resolution in Single-Slice Structured Illumination Microscopy by Thick Slice Blind-SIM Reconstruction.

Jost A, Tolstik E, Feldmann P, Wicker K, Sentenac A, Heintzmann R - PLoS ONE (2015)

Bottom Line: In this work, we show that a single slice of Structured Illumination Microscopy (two or three beam SIM) data can be processed to provide an image exhibiting tight sectioning and high transverse resolution.Our reconstruction algorithm is adapted from the blind-SIM technique which requires very little knowledge of the illumination patterns.We named this new algorithm thick slice blind-SIM because it models a three-dimensional sample even though only a single two-dimensional plane of focus was measured.

View Article: PubMed Central - PubMed

Affiliation: Institute of Physical Chemistry, Abbe Center of Photonics, Friedrich-Schiller University Jena, Jena, Germany; Leibniz-Institute of Photonic Technology, Jena, Germany.

ABSTRACT
The microscope image of a thick fluorescent sample taken at a given focal plane is plagued by out-of-focus fluorescence and diffraction limited resolution. In this work, we show that a single slice of Structured Illumination Microscopy (two or three beam SIM) data can be processed to provide an image exhibiting tight sectioning and high transverse resolution. Our reconstruction algorithm is adapted from the blind-SIM technique which requires very little knowledge of the illumination patterns. It is thus able to deal with illumination distortions induced by the sample or illumination optics. We named this new algorithm thick slice blind-SIM because it models a three-dimensional sample even though only a single two-dimensional plane of focus was measured.

No MeSH data available.


Related in: MedlinePlus

Experimental results in an MCF-7 actin-labelled cell.A 200×200 pixels region of interest was selected to keep the computational time small. a) 2D wide-field (WF) deconvolution. b) 2D blind-SIM reconstruction with higher resolution but no optical sectioning. c) 3D WF deconvolution. d) Thick slice blind-SIM result. The resolution is improved and the out-of-focus contribution removed. The green arrows indicate a pair of filaments that is removed because it originally stems from another plane. The processing time was 25 min with GPU vs. 330 min without GPU. e) and f) are the reconstructions of the original 3D data with the ZEN software (version 2010D). e) Plane that was selected. f) Two slices under slice e), i.e. 182 nm away. The sample was prepared by Michael Reuter and data acquired by Elen Tolstik on a commercial ELYRA-S.1 SIM microscope (Carl Zeiss Microimaging, Jena, Germany). Scale bar: 2 μm.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4493150&req=5

pone.0132174.g003: Experimental results in an MCF-7 actin-labelled cell.A 200×200 pixels region of interest was selected to keep the computational time small. a) 2D wide-field (WF) deconvolution. b) 2D blind-SIM reconstruction with higher resolution but no optical sectioning. c) 3D WF deconvolution. d) Thick slice blind-SIM result. The resolution is improved and the out-of-focus contribution removed. The green arrows indicate a pair of filaments that is removed because it originally stems from another plane. The processing time was 25 min with GPU vs. 330 min without GPU. e) and f) are the reconstructions of the original 3D data with the ZEN software (version 2010D). e) Plane that was selected. f) Two slices under slice e), i.e. 182 nm away. The sample was prepared by Michael Reuter and data acquired by Elen Tolstik on a commercial ELYRA-S.1 SIM microscope (Carl Zeiss Microimaging, Jena, Germany). Scale bar: 2 μm.

Mentions: The novelty of ourthick sliceimplementation is that, even though 2D data were acquired, a 3D deconvolution is performed. The algorithm calculates with a full three-dimensional volume of P focal slices, but in the comparison step between the 3D data predicted from the 3D guess and the 2D measured data, only the central slice is computed and all other comparison values are simply set to zero at this step. The whole stack is then propagated back from measurement space into object space as typically done for gradient-based maximum-likelihood methods [13]. The algorithm minimizes Eq (2), with h being the 3D PSF. We therefore perform a three dimensional deconvolution and the object estimates are three dimensional despite the data being two-dimensional in nature. Due to the axial extent of the PSF, information—the out-of-focus contribution present in the focal slice—will propagate to and accumulate in the extra planes during the deconvolution process and the initially empty planes in 3D sample estimate are filled. The error is calculated as described above in the middle slice only, without other planes contributing. Thus, the algorithm is free to place any information into these extra planes with the goal to decrease the error in the middle plane. The user can choose the number of reconstructed planes P according to the sampling of the discretized PSF, denoted scZ in Table 1. P should be large enough to enable out-of-focus contribution to be efficiently rejected. However, increasing it further unnecessarily increases the computational time.


Optical Sectioning and High Resolution in Single-Slice Structured Illumination Microscopy by Thick Slice Blind-SIM Reconstruction.

Jost A, Tolstik E, Feldmann P, Wicker K, Sentenac A, Heintzmann R - PLoS ONE (2015)

Experimental results in an MCF-7 actin-labelled cell.A 200×200 pixels region of interest was selected to keep the computational time small. a) 2D wide-field (WF) deconvolution. b) 2D blind-SIM reconstruction with higher resolution but no optical sectioning. c) 3D WF deconvolution. d) Thick slice blind-SIM result. The resolution is improved and the out-of-focus contribution removed. The green arrows indicate a pair of filaments that is removed because it originally stems from another plane. The processing time was 25 min with GPU vs. 330 min without GPU. e) and f) are the reconstructions of the original 3D data with the ZEN software (version 2010D). e) Plane that was selected. f) Two slices under slice e), i.e. 182 nm away. The sample was prepared by Michael Reuter and data acquired by Elen Tolstik on a commercial ELYRA-S.1 SIM microscope (Carl Zeiss Microimaging, Jena, Germany). Scale bar: 2 μm.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4493150&req=5

pone.0132174.g003: Experimental results in an MCF-7 actin-labelled cell.A 200×200 pixels region of interest was selected to keep the computational time small. a) 2D wide-field (WF) deconvolution. b) 2D blind-SIM reconstruction with higher resolution but no optical sectioning. c) 3D WF deconvolution. d) Thick slice blind-SIM result. The resolution is improved and the out-of-focus contribution removed. The green arrows indicate a pair of filaments that is removed because it originally stems from another plane. The processing time was 25 min with GPU vs. 330 min without GPU. e) and f) are the reconstructions of the original 3D data with the ZEN software (version 2010D). e) Plane that was selected. f) Two slices under slice e), i.e. 182 nm away. The sample was prepared by Michael Reuter and data acquired by Elen Tolstik on a commercial ELYRA-S.1 SIM microscope (Carl Zeiss Microimaging, Jena, Germany). Scale bar: 2 μm.
Mentions: The novelty of ourthick sliceimplementation is that, even though 2D data were acquired, a 3D deconvolution is performed. The algorithm calculates with a full three-dimensional volume of P focal slices, but in the comparison step between the 3D data predicted from the 3D guess and the 2D measured data, only the central slice is computed and all other comparison values are simply set to zero at this step. The whole stack is then propagated back from measurement space into object space as typically done for gradient-based maximum-likelihood methods [13]. The algorithm minimizes Eq (2), with h being the 3D PSF. We therefore perform a three dimensional deconvolution and the object estimates are three dimensional despite the data being two-dimensional in nature. Due to the axial extent of the PSF, information—the out-of-focus contribution present in the focal slice—will propagate to and accumulate in the extra planes during the deconvolution process and the initially empty planes in 3D sample estimate are filled. The error is calculated as described above in the middle slice only, without other planes contributing. Thus, the algorithm is free to place any information into these extra planes with the goal to decrease the error in the middle plane. The user can choose the number of reconstructed planes P according to the sampling of the discretized PSF, denoted scZ in Table 1. P should be large enough to enable out-of-focus contribution to be efficiently rejected. However, increasing it further unnecessarily increases the computational time.

Bottom Line: In this work, we show that a single slice of Structured Illumination Microscopy (two or three beam SIM) data can be processed to provide an image exhibiting tight sectioning and high transverse resolution.Our reconstruction algorithm is adapted from the blind-SIM technique which requires very little knowledge of the illumination patterns.We named this new algorithm thick slice blind-SIM because it models a three-dimensional sample even though only a single two-dimensional plane of focus was measured.

View Article: PubMed Central - PubMed

Affiliation: Institute of Physical Chemistry, Abbe Center of Photonics, Friedrich-Schiller University Jena, Jena, Germany; Leibniz-Institute of Photonic Technology, Jena, Germany.

ABSTRACT
The microscope image of a thick fluorescent sample taken at a given focal plane is plagued by out-of-focus fluorescence and diffraction limited resolution. In this work, we show that a single slice of Structured Illumination Microscopy (two or three beam SIM) data can be processed to provide an image exhibiting tight sectioning and high transverse resolution. Our reconstruction algorithm is adapted from the blind-SIM technique which requires very little knowledge of the illumination patterns. It is thus able to deal with illumination distortions induced by the sample or illumination optics. We named this new algorithm thick slice blind-SIM because it models a three-dimensional sample even though only a single two-dimensional plane of focus was measured.

No MeSH data available.


Related in: MedlinePlus