The problem — Light-sheet and expansion-microscopy experiments produce multi-tile, multi-view volumes that cannot simply be placed side by side. Tiles overlap imperfectly; dual-illumination views are acquired from opposing angles and must be fused. Without sub-pixel registration, every downstream measurement inherits a spatial error that no amount of segmentation sophistication can correct.
What it is / how it works — BigStitcher is an open-source Fiji plugin built on ImgLib2 that ingests arbitrarily large (terabyte-scale) tiled and multiview acquisitions and solves the alignment problem in three steps: pairwise feature-based registration, global optimisation of the resulting tile graph, and spatially resolved quality estimation that flags poorly aligned regions before fusion. Interactive BigDataViewer previews let an analyst inspect intermediate results at any pyramid level without loading the full volume into RAM. Dual-illumination views are fused and optionally deconvolved in real time on GPU or CPU. All steps are scriptable via the ImageJ macro language and can be run headlessly in CI or HPC pipelines.
Where it fits in the Fovea workflow — BigStitcher is the stitching stage for light-sheet pipelines. It sits between ingest (where PetaKit5D handles raw I/O at speed) and downstream segmentation and tracking (where TrackMate or ultrack consume the fused volume). A registration error here propagates silently into every subsequent coordinate — it is one of the stages the pipeline note identifies as "where the biology is made or lost."
Where it breaks — BigStitcher assumes sufficient texture contrast for feature detection; isotropic, sparsely labelled samples or very thick cleared tissue with refractive-index gradients can yield degenerate feature matches. The spatially resolved quality map is a useful diagnostic, but it does not automatically retry with alternative parameters — that judgement still requires a trained analyst.