Spine Localization and Classification

../../_images/SpineLocalizationAndClassification.png

Summary

Detects, segments and classifies vertebrae in a calibrated CT or MRI volume. Use this to obtain a SpineData with vertebra positions, anatomical types (e.g., cervical/thoracic/lumbar) and levels, label maps and / or surface meshes of vertebrae, illium, sacrum and vertebral discs (MRI only).

Input

  • Single-volume SharedImageSet (CT/MRI), multi-volume inputs are not supported. Ensure the modality is correctly set for the input; otherwise modality can be set manually.

  • Optionally a label image or existing SpineData can be provided.

Output

  • SpineData with vertebra entries containing surface mesh, body center and pedicle keypoints, and vertebra label maps.

Controller capabilities

  • Model selection: choose model set/version if needed. This can be used to select the preferred Machine Learning engine and the set (version) of machine learning models.

  • Export: save state (including current SpineData), export label maps and segmentation images after segmentation.

Workflow

  1. Load a CT or MRI image and (optionally) define a region of interest/bounds.

  2. Localize Vertebrae: predict bounds, set bounds, detect vertebra landmarks, and classify types to create initial SpineData entries.

  3. (Optional) Edit vertebrae: - Add Vertebra: place body/left/right keypoints to add a vertebra. - Remove Vertebra: delete selected vertebra; classification updates accordingly. - Pin/Correct Types: pin type for selected vertebra; re-run classification.

  4. (Optional) Segment structures: - Segment Vertebrae: segment all vertebrae; review meshes. - Segment Sacrum and Ilium: segment pelvis structures where supported. - Segment Discs (MRI only): run disc segmentation for MRI inputs. - Export Label Map: export a multi-label volume assembled from current segmentations.

  5. (Optional) Context and masking: - Run Context Model: compute anatomical context predictions and add them to the data model when available. This aids the classification model in the assignment of anatomical types. - Apply Mask: create/apply a spine mask to the original source image.

  6. (Optional) Save/Load: persist or restore the algorithm state (includes SpineData). See Save and load.