Abstract:

This is a recording of a presentation from a bioimaging workshop focused on Python and the napari image viewer, held as part of the Libre Hub project for disseminating open-source tools in Latin America. The speaker discusses two napari plugins: napari-animation for creating videos and a-finder for manual image registration.

The napari-animation segment demonstrates how to create data tours and visualize analysis pipelines. The plugin works by capturing keyframes of the viewer's state (camera angle, visible layers, etc.) and interpolating between them. Its main advantage, as far as I can tell, is its ability to generate smooth videos from very large datasets that are otherwise too slow to interact with smoothly, as it renders each frame independently.

The a-finder segment introduces a tool for interactive, manual image alignment, which is useful when automated methods fail, especially with multi-modal data like SEM and fluorescence microscopy. The workflow involves an initial three-point correspondence to establish a rough alignment, after which the tool assists in placing additional points for refinement by live-updating the image's affine transformation matrix. The speaker notes its current limitations in 3D.

A Workshop on napari Plugins for Bioimaging Analysis

• 0:00 Workshop Context: This is the third day of a workshop teaching Python and napari for bioimaging. The event is part of the Libre Hub project, which aims to promote open-source hardware and software in Latin America.
• 3:39 napari-animation Plugin: The speaker introduces napari-animation, a popular plugin for creating data presentations. It works by capturing the viewer state at different "keyframes" to generate an animation, which is useful for telling a story with data.
• 4:50 Animating Large, Slow Datasets: A key feature is demonstrated with a >200 GB dataset. Even if the data renders too slowly for smooth interactive viewing, the plugin can produce a fluid video by taking the necessary time to render each frame individually.
• 9:11 Programmatic Control: napari-animation can be controlled via a Python script. This allows for creating reproducible animations that showcase complex data explorations, like moving through z-stacks and time points, changing color maps, and rotating the 3D view.
• 17:07 Visualizing Analysis Pipelines: The plugin can be used to visually explain an image analysis workflow. By sequentially making layers visible (e.g., raw data, then segmentation edges, then final labels) and capturing keyframes, it's possible to create a step-by-step video of the process.
• 19:28 a-finder Plugin for Manual Registration: The presentation shifts to a-finder, a plugin for manually aligning images. It is designed for situations where automated algorithms are likely to fail, such as aligning images from different modalities (e.g., SEM and fluorescence).
• 20:44 Alignment Workflow: The process involves selecting a "reference" and "moving" layer. The user places three corresponding points in both images for an initial affine alignment. After this, the tool helps the user quickly add more points to refine the transformation.
• 24:48 Future Work and Limitations: The speaker acknowledges that a-finder's usability in 3D is currently poor due to limitations in napari's 2D view. Future plans include adding an option to use the manual alignment as a starting point for an automated registration algorithm.