News |  Sitemap |  Contact
PDF Export  | 

EMMA-Box a modular plugin for space variant PSF deconvolution

Abstract

3D Fluorescence microscopy has become an essential tool for biological research especially for functional studies. Unfortunately the acquired images suffer from blur caused by the image formation process itself as the acquired image is the result of the real object image convolved with the system transfer function also known as point spread function (PSF). The PSF usually presents spatial variation mainly along the optical axis (Z axis) rendering the deconvolution process computationally heavy.

Fast classical deconvolution algorithms assume the space invariant of the PSF in order to use a convolution in Fourier space, however such assumption may lead to large errors in the estimation result especially in the case of deep specimens. We propose in this paper a new approach where any classical invariant algorithm can be used with a set of known PSF at various positions to perform a space variant PSF deconvolution. The method is called EMMA and is based on performing multiple invariant deconvolutions and merging the best region of each using an adapted set of masks producing a single estimation.

We describe the EMMA implementation as a highly modular ImageJ plugin, allowing the addition of algorithms and changing the mask form and/or merging process by the mean of “sub-plugin” (plugin of plugin). Actually the algorithms and the mask properties can be written as a plugin by implementing some predefined interfaces. The parameters required for the new function are automatically detected and presented in the EMMA graphical user interface on runtime (Java Reflect).

EMMA process is parallel by nature thereby a multithread processing mode can be applied automatically if the used algorithm allows it (multithread safe). The user can also apply a special approach on iterative algorithms that ameliorates the convergence in EMMA space.

The Meta-EMMA plugin will be soon available on ImageJ website including mask for depth variant PSF and some prewritten algorithms such as LLS and Lucy-Richardson in normal and CUDA accelerated version.

Keywords

deconvolution, PSF, reflection, space variant

Short CV

PHD in Signal processing and automation acquired in 2010, worked as a postdoctoral position for 18 month in MIPS laboratories Haut-Alsace University. Currently I am a research engineer in a collaborative project between Haut-Alsace University and the CEA (Commissariat à l’énergie atomique).

My thesis dealt with 3D deconvolution techniques with space variant PSF, and the Macroscope case was studied during my postdoctoral position. Currently I study the automatic detection and classification of human chromosomal aberrations.

Administrative data

Presenting author: Elie Maalouf
Organisation: MIPS - Université de haute alsace

co-authors: Bruno Colicchio

Alain Dieterlen

© Luxembourg Institute of Science and Technology | Legal Notice