Skip to content

Changelog

1.0.0 — 2026-10-25

First public release.

Portal and API

  • Upload of folders, extensionless DICOM files and ZIP archives through resumable sessions (requests up to 32 MiB); assessment groups images by study, series and modality and reports why a series cannot be used.
  • Versioned automation API under /api/v1 with OpenAPI documentation at /api/docs. One job per selected series; jobs run in a serialized GPU queue with progress, per-task time limits, retries from retained input and deletion.
  • Local GPU and SSH GPU workers, with per-task routing of selected models to an SSH worker.
  • Model catalog built from the installed TotalSegmentator runtime and committed external-model catalogs. Each task exposes its modality, licence, exact exported ROI names, structure kind and the reason it is disabled when it cannot run.
  • Slice viewer for uploaded and completed series with patient orientation markers and RTSTRUCT overlays; prompted models take points, boxes or text from the viewer.
  • MRI sequence roles (for example T2W, ADC and high-b-value DWI) selected from standard DICOM evidence for multi-sequence prostate models.
  • NV-Reason-CT research draft reports for covered chest and abdomen CT, and a CPU crop tool for prostate SBRT CT and RTSTRUCT.

RTSTRUCT export

  • Masks are mapped directly onto the source image grid. Export checks source SOP references, physical bounds and a contour-to-mask round trip (Dice ≥ 0.95, plane tolerance 0.05 voxel) before an RTSTRUCT is released.
  • A shared, lossless orientation contract converts DICOM to NIfTI for every model and restores predictions to the acquired grid; external tasks are enabled only with recorded orientation evidence.

Result manifest and client

  • Every segmentation result ZIP contains rtsegmentator_manifest.json (schema rtsegmentator.result-manifest 1.0, published in schemas/result-manifest-1.0.schema.json). It lists each file with its size and SHA-256, the source series, and per model the outcome, the exact ROI names written, the catalog label contract, and the catalog's licence and limitation text where recorded. GET /api/v1/jobs/{id}/manifest returns it for completed jobs.
  • rtseg_client.py, a standard-library Python client and command-line tool, uploads series, submits jobs, waits, streams results to disk and verifies every package against its manifest and against the manifest the portal serves for the job. It never replays a job submission after an ambiguous failure, refuses redirects to another origin so proxy credentials stay with the portal, and rejects noncanonical or colliding member names. RTpipeline vendors the same file to import RTsegmentator results as a segmentation source.

Distribution

  • Citation metadata (CITATION.cff, codemeta.json, .zenodo.json), a documentation website built with MkDocs, and a generated model catalog page.
  • The Apache-2.0 licence covers original RTsegmentator source code only. Model weights, runtimes and datasets are not distributed and remain under their own licences.