Lexicon

RadiObject-specific terminology. For general medical imaging terms, see the DICOM standard, NIfTI specification, and BIDS specification.

Coordinate Systems and Orientation

Term	Definition
RAS	Right-Anterior-Superior — standard neuroimaging coordinate system (NIfTI default)
LPS	Left-Posterior-Superior — ITK/some DICOM coordinate system
axcodes	3-character tuple indicating axis directions, e.g., `('R', 'A', 'S')`
affine matrix	4x4 matrix mapping voxel indices `(i,j,k)` to world coordinates `(x,y,z)`
axial / sagittal / coronal	Anatomical planes: top-down, side, and front views respectively

Term	Definition
RadiObject	Top-level TileDB Group organizing subject metadata (ObsMeta) and multiple VolumeCollections. Filtering returns views (`is_view=True`).
is_view	Boolean property indicating whether a RadiObject/VolumeCollection is a filtered view
obs_subject_id	Unique string identifier for a subject. Primary key in `obs_meta`, foreign key in `obs`
ObsMeta	Subject-level observational metadata. Single-dimension TileDB sparse array keyed by obs_subject_id only. Contains system-managed `obs_ids` column (JSON list of volume obs_ids per subject).
VolumeCollection	TileDB Group organizing multiple Volumes with consistent X/Y/Z dimensions. Filtering returns views.
Volume	Single 3D or 4D radiology acquisition backed by TileDB dense array
obs_id	Unique string identifier for a Volume, unique across entire RadiObject
obs	Volume-level observation dataframe containing per-volume metadata
Dataframe	TileDB-backed sparse array with configurable index dimensions for tabular data
INDEX_COLUMNS	Default dimension tuple `(obs_subject_id, obs_id)` for VolumeCollection.obs. obs_meta uses `(obs_subject_id,)` only.
obs_ids	System-managed obs_meta attribute: JSON-serialized sorted list of volume obs_ids linked to each subject
generate_obs_id()	Creates obs_id from `f"{obs_subject_id}_{collection_name}"`. Ensures global uniqueness across collections

Term	Definition
Index	Immutable named dataclass providing bidirectional mapping between string IDs and integer positions. Supports set algebra (`&`, `\\|`, `-`, `^`), alignment checking (`is_aligned`).
align()	Standalone function computing the intersection of multiple Index objects
iloc	Integer-location based indexer for selecting by position
loc	Label-based indexer for selecting by obs_subject_id or obs_id
sel()	Named-parameter selection by obs_subject_id
subjects	VolumeCollection property returning deduplicated subject IDs in insertion order
groupby_subject()	VolumeCollection iterator yielding `(subject_id, VolumeCollection)` pairs
filter()	Filter subjects using TileDB QueryCondition expression
head() / tail() / sample()	Return view of first/last/random n subjects
select_collections()	Filter RadiObject to specified collections (returns view)
describe()	Return summary showing subjects, collections, shapes, and label distributions

Term	Definition
LazyQuery	Lazy filter builder for VolumeCollection with deferred transforms. Entry point: `vc.lazy()`
EagerQuery	Computed query results with pipeline methods (`.map()`, `.write()`, `.to_list()`). Entry point: `vc.map(fn)`
lazy()	VolumeCollection method returning LazyQuery for deferred transforms
map()	Apply transform to each volume. On VolumeCollection returns EagerQuery (eager); on LazyQuery defers execution
write()	Persist query results to new VolumeCollection
to_list()	Extract raw results from EagerQuery
to_dataframe()	Summarize EagerQuery as a DataFrame with obs metadata and a `result` summary column. Accepts optional `columns` to override obs column selection
iter_volumes()	LazyQuery streaming iterator yielding Volume objects
to_numpy_stack()	Load all matching volumes as stacked array `(N, X, Y, Z)`
TransformFn	Type alias: `Callable[[np.ndarray, pd.Series], TransformResult]` — receives `(volume, obs_row)`
TransformResult	Union type: `np.ndarray` or `tuple[np.ndarray, dict[str, AttrValue]]` — volume alone or volume + obs updates
BatchTransformFn	Type alias for batch transform: receives list of `(volume, obs_row)` tuples

Term	Definition
VolumeCollectionWriter	Context manager for incremental VolumeCollection writes
RadiObjectWriter	Context manager for building RadiObject incrementally
write_volume()	Write single volume data to VolumeCollectionWriter
write_batch()	Write multiple volumes at once
add_collection()	Add new collection to RadiObjectWriter, returns VolumeCollectionWriter
write()	Persist a RadiObject or VolumeCollection (including views) to storage

Term	Definition
from_images	Factory method for bulk ingestion (NIfTI and DICOM) via `images` dict with auto format detection
images	`from_images` param: dict mapping collection names to image sources (glob, directory, or pre-resolved tuples)
ImageFormat	Enum (`NIFTI`, `DICOM`) representing detected or hinted image format
format_hint	Optional dict mapping collection names to format strings (`"nifti"` or `"dicom"`) for ambiguous sources
validate_alignment	Verify all collections have matching subject IDs
series_type	BIDS-aligned identifier for imaging sequence type (T1w, T2w, FLAIR, bold, dwi, CT)
append()	Atomic method to add new subjects to existing RadiObject or VolumeCollection

Term	Definition
partial read	Reading a subset of volume data without loading the entire volume
axial() / sagittal() / coronal()	Partial read returning a 2D plane at specified coordinate
slice()	Partial read returning arbitrary 3D ROI
to_numpy()	Full volume read returning complete numpy array
to_nifti()	Export volume to NIfTI file format
get_statistics()	Compute descriptive statistics (mean, std, min, max, percentiles)
compute_histogram()	Compute intensity histogram

Term	Definition
OrientationInfo	Pydantic model storing anatomical orientation metadata
tile_orientation	Volume property returning SliceOrientation enum for tile chunking strategy
canonical orientation	Standard RAS orientation where X+=Right, Y+=Anterior, Z+=Superior
reorientation	Transforming volume data and affine to match a target orientation

Term	Definition
RadiObjectConfig	Top-level configuration with nested `write`, `read`, and `s3` settings
WriteConfig	Settings applied when creating new TileDB arrays. Immutable after creation
ReadConfig	Read settings (memory_budget_mb, concurrency, max_workers)
TileConfig	Tile chunking configuration (orientation, extents)
CompressionConfig	Compression settings (algorithm, level)
OrientationConfig	Anatomical orientation settings during ingestion
S3Config	Cloud storage settings (region, endpoint, max_parallel_ops)
configure()	Global function to update configuration
get_tiledb_ctx()	Returns lazily-built TileDB context from current configuration

Term	Definition
VolumeCollectionDataset	PyTorch Dataset wrapping VolumeCollection(s)
VolumeCollectionSubjectsDataset	TorchIO-compatible Dataset yielding `tio.Subject` objects
LoadingMode	Enum: `FULL_VOLUME`, `PATCH`, or `SLICE_2D`
DatasetConfig	Pydantic model configuring dataset behavior
create_training_dataloader	Factory function producing configured DataLoader

Term	Definition
ctx_for_threads()	Returns shared TileDB context for thread pool workers (metadata caching works)
ctx_for_process()	Creates new TileDB context for forked processes (DataLoader workers)
persistent_workers	DataLoader setting to keep worker processes alive between epochs
pin_memory	DataLoader setting to copy tensors into CUDA pinned memory