01 · Case study

Multistage AI pipeline for spine segmentation

A two-stage pipeline (Stage 1 · Initial segmentation, Stage 2 · Vertebra Detection) delivered as a 3D Slicer module, with Body, Body + Pedicles and Center of mass outputs.

Surgical Works · 3D Slicer module · CT

Pipeline visual

From CT volume to labeled spine outputs

3D Slicer module
Stage 2Vertebra detection3D Volume: coronal CTStage 1: initial segmentation of spine and pelvisStage 2: vertebra detection
3D outputs
Body 3D outputBody
Body + Pedicles 3D outputBody + Pedicles
Center of mass 3D outputCenter of mass
  1. 013D VolumeInput · coronal CT
  2. 02Stage 1Initial segmentation
  3. 03Stage 2Vertebra detection
  4. 04OutputsThree 3D results
The highlighted step is the current pipeline stage.
Model evidence

Expert and model comparison

Reference and model views shown side by side for review.

Expert · liver segmentation
Expert · liver segmentation
Model · liver segmentation
Model · liver segmentation
Expert · lesion segmentation
Expert · lesion segmentation
Model · lesion segmentation
Model · lesion segmentation
The task

Segment the full spine on CT and identify each vertebra.

What we built

A multistage AI pipeline running as a 3D Slicer module.

The result

Three 3D outputs per case: Body, Body + Pedicles, Center of mass.

Workflow

CTStage 1Stage 23D model in Slicer

Input
3D CT volume
Stages
Initial segmentation → Vertebra Detection
Outputs
Body · Body + Pedicles · Center of mass
Delivered as
3D Slicer module (volume: CadavericVolume; segments: Cervical, ToracoLumbar, Pelvis)
  • Re-export six stage/output stills at 2x from original data; outputs need transparent background.
Discuss this projectNext: End-to-end full-spine segmentation in OHIF →