Annotation expert
Check windowing and anatomy before accepting the AI contour for Skull. Then revise pixels, continuity, and negatives.
- Draft vs final kept apart
- Disputes go to multi-blind review
- Export only reviewed results
“Skull” is a preset AI auto-label model in the Langhui medical annotation platform. Default modality: CT. Region: Bone. Type: Organ / structure. Batch inference outputs Mask or Contour drafts; drafts enter training sets only after revision and blind review. This page describes model capability and workflow — not a disease product and not in-stock case volume.
Bone imaging tasks usually cover vertebrae, ribs, and joints. Attachment note: Preset model in the Langhui medical annotation platform. Triggers AI auto-labeling.
One system, three acceptance questions. Answers must map to the status machine.
Check windowing and anatomy before accepting the AI contour for Skull. Then revise pixels, continuity, and negatives.
Reproducibility first: DICOM in, model version v1.0, Mask/Contour out, MPR continuity, review status on labels.
Match task to Bone / CT, then confirm whether data leaves the hospital network. No diagnostic claims. No company-wide accuracy.
Same main path: manage → AI pre-label → refine → review → export.
Platform default path remains CT / MR / PET · DICOM. This list modality is CT; US/MG entries need project assessment. Endoscopy video, pathology WSI, and clinical NLP are out of default scope.
Task assignment and status (to-annotate → annotated → pending review → approved) stay visible. AI output for Skull is a draft by default.
High-risk or lesion tasks can force final review. Dice/Kappa rules are defined by the project — not a company-wide accuracy score.
After sample review confirms fit, agree batch, license, and export formats. Public pages do not provide diagnosis or registration promises.
“Skull” is a preset AI auto-label model name. The platform has 186 preset models, not 186 disease products. Project fit depends on task definition, authorization, and the data card.
No. Model output is a draft. Revision plus single- or multi-blind review is required before export. Pre-labeling is not diagnosis and does not promise sensitivity or Dice.
Exports should align images, contours, guideline version, model version (v1.0), and review status. Use MPR for 3D continuity. Version changes go to the next dataset freeze.
Confirm whether data leaves the network, upload de-identification, and permissions. Then pilot: ingest → trigger Skull → revise → blind review → export. YY/T 1833.3 is educational alignment only.
Bring anatomy, modality, and on-prem needs to request a Skull sample review.