Knee MRI Segmentation
nnU-Net cascade — automatic segmentation of 9 knee structures (patellar / femoral / tibial cartilage, medial & lateral meniscus, femur, tibia, patella) with per-structure volumes, cartilage thickness and meniscus morphometry.
Research tool — not a regulated medical device. Their results are not a diagnosis and can be wrong. Every result must be reviewed by a qualified clinician, who remains responsible for the clinical decision.
Source: nnU-Net
This model is heavy (GPU, ~1–2 min) and runs from the PACS worklist, not as a browser upload. Open a knee MRI study, choose AI → Knee analysis, and a labelled segmentation overlay appears in the OHIF viewer with a morphometry PDF attached to the study.
What it does
Segments cartilage, meniscus and bone on a knee MRI, then computes per-structure volumes, cartilage thickness statistics and meniscus height/volume. The overlay is a standard DICOM-SEG — it reformats to all planes and 3D in the viewer, exactly like the brain-tumor and spine modules.
Sagittal knee MRI with the 9-structure AI overlay — drag to scroll slices, switch to 3D for the surfaces
Morphometry (example)
Femoral cartilage volume
7.46 mL
Medial meniscus
2.31 mL
Femoral cartilage thickness
1.1 mm
median 1.0 · p95 1.9
Segmented structures
Volumes measured on the example above — the same numbers the attached PDF report carries.
Report metrics
- Volumes — per-structure volume (mL) for all 9 structures.
- Cartilage thickness — mean / std / median / p5 / p95 (mm), measured across the cartilage layer: from the bone-cartilage interface to the articular surface. Research metric, not a validated clinical thickness.
- Meniscus — height (mm) and volume (mL) for the medial and lateral meniscus.
How to run it
- Open the PACS worklist and find a knee MRI study.
- Click the AI menu on that row and choose Knee analysis.
- After a couple of minutes the labelled overlay appears in the OHIF Segmentation panel and a morphometry PDF is attached to the study.
Model: nnU-Net v2 cascade (Dataset500_KneeMRI, OAI-ZIB). Mean Dice 0.906 across 9 structures. Goyal et al., 2025.