Inspect Coverage and Label Loss in a Cortical Projection

Returns diagnostics computed by apply_surface_projection(). Label counts are separated by map and stage, so a key surviving another map cannot hide loss. not_sampled means a source key has no supported sampled vertex; lost_resampling means it survives sampling but disappears during surface resampling. Declared keys absent from the source have status absent_source. Opposite-hemisphere keys have status other_hemisphere and are excluded from expected parcel-loss totals; any supported output vertices carrying those keys are still counted as hemisphere mismatches. Without explicit hemisphere metadata, keys remain unchecked and no hemisphere is inferred. Key zero is counted separately and excluded from parcel-loss totals; supported zero is still a valid value. Diagnostics never mask, relabel or repair values.

Usage

projection_diagnostics(x)

Arguments

x

A SurfaceData result from apply_surface_projection().

Value

A list with summary, a per-map coverage/count tibble, and labels, a per-map/per-key tibble for label data (NULL for other data types). Label rows contain source voxel counts, supported sampled and target vertex counts, hemisphere declarations, loss flags and stage status. Vertex and voxel counts are not interchangeable area or volume measures. Missing and masked vertices are excluded from label counts and retained in coverage.

Examples

if (FALSE) { # \dontrun{
qa <- projection_diagnostics(projected_atlas)
qa$summary
subset(qa$labels,
  lost_at_sampling | lost_at_resampling | hemisphere_mismatch
)
} # }