Data structures and coordinate conventions¶
Collection, image, and cell¶
ImageCollection holds a dataset path and calibration, loaded phase images,
masks, channel dictionaries, image objects, geometry tables, feature tables,
and processing records. Image holds a single field and its cells, channel
data, background/interpolation caches, and stage-specific tables. Cell holds
one cell’s contour, midline, mesh, detected-object data, feature dictionaries,
and discard state.
Object |
Shape / convention |
|---|---|
Image and mask |
|
Contour and midline |
|
Cell mesh |
|
Object geometry |
Lists of contours, meshes, and midlines per channel; failed meshes may be empty arrays. |
Profile feature |
A one-dimensional array or list stored in a dataframe cell. |
The historical mesh attributes x1, y1, x2, and y2 are slices of
the stored geometry. OpenCV drawing needs Cartesian x/y points, so the curated
mask exporter explicitly reverses contour columns. Plot overlays also adapt
the convention. Do not reverse arrays solely because an attribute says x.
Object dictionary¶
cell.object_meshdata[channel] contains detected object_contour,
object_mesh, and object_midline lists, with construction-error information
where applicable. Their entries remain aligned even when a mesh fails.
Feature dictionaries¶
The methods populate morphological_features, profiling_features,
objects_features, membrane_features, and svm_features on cells.
profiling_data stores intermediate data, including widths used during
morphology, and is not automatically another exported feature family.
Collection tables¶
mesh_df_collection stores combined geometry. feature_dataframes holds
separate family/channel tables. merged_features is created by merging;
it is invalidated by later feature computation or image population recreation.
Do not export an earlier merged table after changing the calculation state.
cell_id and frame are analysis identities. They are not lineage IDs or
evidence that the package tracks cells through time.