Pretrained multi-organ segmentation

Run TotalSegmentator online, then edit the masks

MedSeg integrates TotalSegmentator into a browser workspace so CT and MRI masks can be generated on hosted GPUs, opened in the editor, corrected, measured, and exported.

117 CT structures 50 MR structures Specialized subtasks
TotalSegmentator-style multi-organ masks displayed in MedSeg.
TotalSegmentator masks in MedSeg become normal editable multi-label segmentations.

Common TotalSegmentator workflows

Use the broad task when you need many structures. Use a subtask when you need a focused target that the broad model does not cover.

Total CT

Whole-body CT segmentation for organs, vessels, skeleton, muscles, and many common anatomical structures.

Total MR

MRI organ segmentation for research workflows where an organ-by-organ baseline mask is useful.

Focused subtasks

Liver lesions, Couinaud segments, lung vessels, effusions, hip implants, head and neck structures, and other targeted tasks.

How to run it in MedSeg

  1. Upload DICOM/NIfTI or copy a public case.
    Keep the original series in the project so masks remain traceable.
  2. Select one or more rows.
    Batch select a cohort if you want the same task run across multiple scans.
  3. Choose TotalSegmentator from Models.
    Pick Total, Total MR, or the relevant subtask in the run dialog.
  4. Review the mask in the editor.
    Inspect labels, correct failures with manual tools or nnInteractive, and calculate volumes.

After inference

The mask is not trapped in the model output folder. It becomes part of the project.

Edit

Lock labels, correct boundaries, merge masks, or remove labels you do not need.

Measure

Use volume analysis and per-label measurements to find outliers across the dataset.

Export

Download masks as NIfTI or build DICOM-compatible bundles for review workflows.

Know when not to use it

TotalSegmentator is not a final answer for every target. For tumors outside explicit lesion subtasks, rare anatomy, pediatric anatomy, or site-specific structures, use nnInteractive, VoxTell, or custom nnU-Net training.