Adaptive multiorientation resolution analysis of complex filamentous network images

MOTIVATION Microscopy images of cytoskeletal, nucleoskeletal, and other structures contain complex junctions of overlapping filaments with arbitrary geometry. Yet state-of-the-art algorithms generally perform single orientation analysis to segment these structures, resulting in gaps near junctions, or assume particular junction geometries to detect them. RESULTS We developed a fully automated image analysis approach to address the challenge of determining the number of orientations and their values at each point in space in order to detect both lines and their junctions. Our approach does not assume any fixed number of orientations or any particular geometry in the case of multiple coincident orientations. It is based on analytically resolving coincident orientations revealed by steerable ridge filtering in an adaptive manner that balances orientation resolution and spatial localization. Combining this multi-orientation resolution information with a generalization of the concept of non-maximum suppression allowed us to then identify the centers of lines and their junctions in an image. We validated our approach using a wide array of synthetic junctions and by comparison to manual segmentation, and applied it to light microscopy images of cytoskeletal and nucleoskeletal networks. AVAILABILITY https://github.com/mkitti/AdaptiveResolutionOrientationSpace. SUPPLEMENTARY INFORMATION Supplementary information is available at Bioinformatics online.