Automatic Segmentation of Wrist Bones in CT Using a Statistical Wrist Shape + Pose Model.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Wrist bone segmentation in CT images is crucial for clinical applications such as arthritis evaluation, bone age assessment, and image-guided interventions.
• Challenges include bone tissue non-uniformity, spongy textures, and narrow inter-bone spaces, hindering accurate segmentation.

Purpose of the Study:

• To develop an automatic wrist bone segmentation technique for CT images.
• To create a statistical model capturing shape and pose variations of the wrist joint.

Main Methods:

• A statistical model was built using 60 example wrists across nine positions, incorporating shape and pose variations.
• Joint alignment via group-wise registration (Gaussian Mixture Model) and Principal Component Analysis (PCA) for shape variations.
• Expectation-maximization and non-rigid registration were used for segmenting wrist bones in CT images.

Main Results:

• The proposed method achieved a mean surface distance error of 0.33 mm.
• A mean Jaccard index of 0.86 was reported, indicating high segmentation accuracy.
• Validation was performed on 66 unseen CT volumes with an average voxel size of 0.38 mm.

Conclusions:

• The developed statistical model and registration approach provide an effective automatic segmentation of wrist bones in CT images.
• This technique holds promise for enhancing various clinical applications requiring precise wrist bone analysis.