HIGH DENSITY MOLECULE LOCALIZATION

Gazagnes, S., Soubies, E., & Blanc-Feraud, L.  High density molecule localization for  super-resolution  microscopy  using  CEL0  based  sparse  approximation.   In  2017  IEEE  14th International Symposium on Biomedical Imaging (ISBI 2017) (pp.  28-31).  

Link to Paper: High density molecule localization

The charactistics Airy disk of two single molecules seen through a perfect optical system.

​From March to August 2016, I worked in the Morpheme Team​ (https://team.inria.fr/morpheme/), which is part of the Inria laboratory I3S, under the supervision of Emmanuel Soubies and Laure Blanc-Féraud.  My project was related to the implementation of a new algorithm to perform high-precision molecule localization using photo-activated localization microscopy techniques (PALM). Typically, conventional microscopy techniques are limited in resolution by the diffraction limit, such that the image of a point through an acquisition system is a spot called an Airy disk. The size of this spot depends on the characteristics of the microscope and the wavelength of the light. If two molecules are too close to each other, we might resolve only a single Airy disk, and assume that we only see a single object. 
​
​
PALM microcopy techniques solve this limitation by sequentially activating and imaging a small set of fluorescent molecules. This requires to use "photoactivable" fluorescent molecules, that we can turn off or on. The idea is to acquire several images with a different set of molecules activated each time. It is then easier to reconstruct the position of each molecule of the different images, and assemble them to reconstruct the entire image. This gives a "super-resolved" image, which can go beyond the diffraction limit of conventional microscopes.

The Photo Activated Localization Microscopy (PALM) approach

---

The quality of the reconstructed image depends on the number of photo-activated molecules in each frame, but also on the quality of  the numerical method used to locate molecules. During these 6 months, I implemented an algorithm which performs the reconstruction of the super-resolved image based on a deconvolution algorithm with a L0 regularization term to promote sparsity.  More details can be found in the paper https://hal.inria.fr/hal-01443565/document.

The code can be found there: https://github.com/esoubies/SMLM-CEL0.  Below are two images showing examples of super-resolved reconstruction using this method. The image in the top left corner is the image obtained using conventional microscopy.

Below, you can find an example of live reconstruction using a different data set.

In short:

• PALM microscopy is a super-resolution technique which can go beyond the resolution limit of conventional microscopy techniques using a sequential activation and image acquisition of fluorescent molecules. 
• The SMLM-CEL0 algorithm is a Matlab package I developed in collaboration with Emmanuel Soubies and Laure Blanc-Feraud to reconstruct the super-resolved image based on the set of individual images.