Joints Analysis¶
Joint space width
Subchondral bone
Articular surfaces
Articular cartilage
Osteochondral junction
Example: Joint space width¶
Create binary masks of the bone to analyze.
If required, use the interactive rotated VOI task to roughly select the joint space area.
Use the Lucid.Bone.JointSpaceWidth to create two masks.
Use the Lucid.Core.Thickness to measure the thickness of the created dilated mask, while limiting the evaluation to the smaller masked region.
This produces various relevant parameters:
JSW.Mean (Mean thickness / joint space width)
JSW.SD (Standard deviation)
JSW.Max (Maximum thickness / joint space width)
JSW.Min (Minimum thickness / joint space width)
JSW.Median (Median thickness / joint space width)
Additional parameters can also be derived, for example:
JSW.AS (Asymmetry) can be calculated using a formula JSW.Max / JSW.Min via using
Lucid.Util.FloatFormula.JSV (Volume) can be calculated using tasktype:`Lucid.Core.CountSegmentsPixels.
There are various alternatives to Lucid.Bone.JointSpaceWidth. Internally it applies a series of morphological operations.
These can instead also be applied explicitly using Lucid.Core.MorphDilate, Lucid.Core.MorphErode etc.
Another alternative is to rely on existing external tools, for example via Example.ORMIRXCT.JointSpaceWidth.
Creating the initial binary masks of the bone can also be done in a variety of ways,
for example using manual interactive segment editing, using morphological operations, or using machine learning task types like Example.nnunet2.SegmentImage.
Mouse Femur / Tibia Segmentation nnU-Net Models¶
We offer models to automatically segment 3D CT images of mouse femur and tibia bone joints. These nnU-Net 2 models were trained for over fifty hours on an RTX 4090. The training data consisted of over 70 images segmented with different techniques, including earlier versions of the models.
The model creates filled whole bone region masks. Those masks can be used in further workflows to be post-processed and combined for example with any thresholding task to automatically mask out for example cortical bone or trabecular bone.
The available models are:
nnunetv2-model-MouseFemurTibiaV1L.zipnnunetv2-model-MouseFemurTibiaV2S.zipnnunetv2-model-MouseFemurTibiaV2L.zip
V2 is the latest version. V2S was trained on smaller images to include more overall shape context, which often provides preferable results. V2L was trained on larger images to include more local details. Post-processing can be beneficial.
The suggested task configuration is:
Input Parameter |
Fixed Value |
|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
See also Machine Learning Models: nnUNet.
