TIFF
2-D TIFF input can enter the current Boundary U-Net segmentation pipeline directly. Shape, dtype and intensity statistics are recorded before inference.
Nikon ND2
ND2 files are read with the maintained nd2 package. Channel, time, Z and field/position selection is explicit for multidimensional acquisitions.
ND2 selection is explicit
ChannelC index
TimeT index
Z / FieldZ + position
Provenance: the source axes, sizes, selected indices, reader version and SHA-256 are recorded for the 2-D plane used by segmentation.
Scalable microscopy roadmap
Broader formats, only after validation.
QuPath demonstrates the value of broad microscopy interoperability and scalable image access. BioNuclei will adopt the useful architecture without claiming support before a reader is tested end to end.
OME-TIFF
Structured metadata and multidimensional acquisition support after reader and provenance tests.
Pyramidal images
Tile based access for large images so inference does not require the complete image in memory.
OME-Zarr
Chunked, scalable image access for larger microscopy datasets and future cloud workflows.
Whole-slide workflows
Region based processing with explicit provenance, QC and aggregation rather than a monolithic upload.
Every new format gets a validation gate.
Reader correctness, axis handling, metadata preservation, deterministic plane selection, memory behaviour and scientific regression tests must pass before a format is exposed in the public analyzer.
Use supported formats