Haar Wavelet Kernel (HAWK) improves High Density 3D Localisation microscopy

Session

Poster Session Two

Authors

Dr Richard Marsh (1), Mr Ishan Costello (1), Dr Susan Cox (1)

Affiliations

1. KCL

Keywords

Localisation Microscopy , 3D-Microscopy, image artefacts, Algorythm

Abstract text

Summary:

Localisation based super-resolution is known to frequently produce image artefacts when the density of emitters is too high. Even purposely designed algorithms suffer bias in their localisations when emitter overlap is sufficiently large. Recently we have developed analysis tools that can both help limit this bias (HAWK) [1] and detect its presence (HAWKMAN) [2] in 2D image reconstructions. Here we show tests of HAWK in various 3D Localisation microscopy modalities. We show both that the nature of artefacts in 3D reconstructions is much more complex and unintuitive than 2D and that HAWK provides a useful tool for reducing these image artefacts.

Introduction:

Localisation microscopy has become a standard technique in recent years, but its use in live cell imaging that requires a high emitter density for data acquisition to be sufficiently fast leads to problems with overlapping single emitters. The normal result of this overlap is that the localisation algorithm will either miss some emitters or gives localisation biased to their mutual centres. This results in an image artefact know as ‘artificial sharpening’ where actual resolving power is reduced but often reported as higher resolution [pat]. These artefacts have been noted and discussed in 2D localisation microscopy [1,2,3] and are to a certain degree intuitive to expert users. We have previously developed an alternative analysis approach (HAWK) that does not suffer the same bias when emitter overlap is high and used this to identify artefacts in 2D localisation images.

Several methods exist for 3D localisation microscopy, usually involving encoding the Z position of emitters into the Point Spread Function. These altered PSFs lead to much more complex emission patterns when two or more emitter PSF overlap. These makes it much more difficult to determine the likely bias in emitter positions when the localisation algorithm fails. Additionally, the nature of the produced artefacts will differ for different 3D modalities and likely be highly dependent on the 3D structure being imaged.

Materials and methods:

Here we test the performance of HAWK to reduce reconstruction artefacts on simulated and experimental data. We explore three different 3D modalities – Astigmatism, Biplane and Double Helix. We use the localisation software ThunderSTORM [4] and EasyDHPSF [5]. We examine the reconstructions both with and without prior HAWK processing for artefacts by comparison with ground truth where possible. Tests are also performed on experimental datasets for diverse structures including the nuclear pore complex [6], Mitochondria.

Results:

We demonstrate that reconstruction artefacts are readily generated in each mode of 3D localisation microscopy and that the nature of the artefacts are much more complex than 2D with the bias not simply just toward the mutual centre. We also show that whilst for Biplane and Astigmatism based 3D artefacts are predominantly produced by biased localisation, Double helix artefacts produced using DHPSF are predominantly produced by missing localisations. In each of these cases the scale of the artefacts produced is substantially reduced by using prior HAWK processing.

Conclusions:

The complex nature of localisation artefacts we find, which differ substantially between 3D localisation modalities suggests they would be very hard to detect even for expert localisation microscopists. The use of HAWK processing to highlight/reduce these artefacts represents a very useful tool in the validation of 3D localisation microscopy images.

 

References

[1]          R. J. Marsh, K. Pfisterer, P. Bennett, L. M. Hirvonen, M. Gautel, G. E. Jones and S. Cox, "Artifact-free high-density localization microscopy analysis," Nature Methods, p. 689–692, 2018.

[2]          Marsh, R.J., Costello, I., Gorey, MA. et al. Sub-diffraction error mapping for localisation microscopy images. Nat Commun 12, 5611 (2021). https://doi.org/10.1038/s41467-021-25812-z

[3]          P. Fox-Roberts, R. Marsh, K. Pfisterer, A. Jayo, M. Parsons and S. Cox, "Local dimensionality determines imaging speed in localization microscopy," Nature Communications, 2017.

[4]          M. Ovesný, P. Křížek, J. Borkovec, Z. Švindrych and G. M. Hagen, "ThunderSTORM: a comprehensive ImageJ plug-in for PALM and STORM data analysis and super-resolution imaging," Bioinformatics, p. 2389–90, 2014.

[5]          easy-dhpsf download | SourceForge.net

[6]          D. Sage, H. Babcock, T.-A. Pham, T. Lukes, T. Pengo, J. Chao, R. Velmurugan, A. Herbert, A. Agrawal, S. Colabrese, A. Wheeler, A. Archetti, B. Rieger, R. Ober, G. M. Hagen, J.-B. Sibarita, J. Ries, R. Henriques, M. Unser and S. Holden, "Super-resolution fight club: assessment of 2D and 3D single-molecule localization microscopy software," Nature Methods, vol. 16, no. 5, pp. 387-95, 2019.