Result format

Every endpoint returns the same JSON object, the result envelope. JSON is the native format; DICOM SR/SEG, FHIR and PDF are renderings of it, requested with options.return.

Top level

Field	Content
sections[]	One entry per section of the run: id, status (complete, not_assessable, not_in_input, not_applicable, failed), reason, standards[], item_ids[]
findings[]	Detected or excluded entities, coded (SNOMED CT), with sites, regions and severity
measurements[]	Numbers with UCUM units, method, uncertainty and the frame or slice they came from
scores[], classifications[], risks[], changes[]	Clinical scores, reporting-system categories, risk predictions and comparisons with priors
triage	critical, urgent, routine or none, with flags pointing to findings
guidance[]	Guideline-based recommendations (guideline and version), marked as decision support
report	Only when options.report is true: the draft narrative, with every printed number bound to an item
artifacts[]	DICOM SR, SEG, Secondary Capture, overlays (PNG/WebM), PDF, FHIR bundle
quality	overall (adequate, limited, non-diagnostic), flags and checks
provenance, timings	Pinned versions, pipeline and components; time per stage
warnings[], errors[]	Workflow problems with codes
extensions.<ns>	Modality-specific data, e.g. extensions.angio for cine runs and key frames

Blocks that do not apply are empty arrays, never missing, so one parser handles every model.

Every number is traceable

Each value carries a trace back to the pipeline output it came from. The draft report binds every number it prints to an envelope item, and the validator rejects the report otherwise.

Codes and standards

Concepts are {system, code, display}: SNOMED CT for findings and anatomy, LOINC for report sections, DICOM codes for templates, UCUM for units. Each section names the standards it follows, with versions.

Your UIDs

Files are processed as received: outputs keep your Study, Series and SOP Instance UIDs, so SEG and SR open next to the originals.