Using steerable wavelets and minimal paths to reconstruct automatically filaments in fluorescence imaging

The accurate detection of filamentous structures in fluorescence microscopy, such as stained cytoskeleton and cilia, is an important technical issue in bioimage analysis. We propose here a two-steps approach that combines image thresholding in steerable wavelets domain and minimal path reconstruction to robustly detect and quantify filaments in their entire length. Indeed, the first steerable transformation enhances bright and anisotropic structures such as stained filaments, but local variations of fluorescence intensity often leads to line breaks in segmented filaments. We thus used a minimal path algorithm in a second step to close these gaps and reconstruct the whole filaments. Thereafter, we used our two-steps approach to detect and quantify the flagellum of the parasite Typanosoma brucei at a population level.