Section: New Results

Multi-modal registration for correlative light-electron microscopy

Participants : Bertha Mayela Toledo Acosta, Patrick Bouthemy, Charles Kervrann.

We pursue our work on correlative light-electron microscopy (CLEM), which combines the strengths of two different imaging modalities, light microscopy (LM) and electron microscopy (EM), to jointly study intracellular dynamics and ultrastructure of a biological sample. CLEM registration is an important and difficult problem given the significant differences between LM and EM images regarding resolution, field of view, image size and appearance. We designed an automated approach for retracing and registering CLEM images, by implementing a patch-based search using a common Laplacian of Gaussian (LoG) image representation of the LM and EM images. We have redefined the geometry of the patch, opting for a disk-shaped patch. The search (or retracing) step uses histogram-based methods as they are invariant to rotation, and it provides a pre-registration by producing the estimate of the translation component. Usually, there is a large disparity on the orientation of EM and LM images. To handle this problem, we have implemented a mutual information-based method to compute the rotation between the EM and LM patches and to refine the registration. We have also tackled the registration issue in both directions (LM to EM, and EM to LM), and compared our approach to a correlation-based method.

We have tested our approach on a larger set of real CLEM images (provided by Institut Curie) presenting a large diversity in content, image size, and appearance, further validating our method (see Fig. 8). We are currently exploring how our automated CLEM registration method could be exploited to guide EM acquisition within a coarse-to-fine framework.

Reference:  [35]

Collaborators: Xavier Heiligenstein (UMR 144 CNRS-Institut Curie),

                          Grégoire Malandain (Inria, Morpheme EPC, Sophia-Antipolis).