BOSS: Bones, organs and skin shape model.

Area of Science:

• Medical imaging
• Computational anatomy
• Biomedical engineering

Background:

• Virtual anatomical models enhance clinical tasks like workflow automation and image-guided interventions.
• Current statistical shape models (SSMs) often lack comprehensive organ/bone representation and population diversity.
• High-quality patient surface and internal organ data are crucial for accurate pose and shape estimation.

Purpose of the Study:

• To develop a deformable human shape and pose model integrating skin, internal organs, and bones.
• To create a holistic anatomical representation for improved medical applications.
• To address limitations of existing SSMs in population representation and scope.

Main Methods:

• A deformable model was learned from CT images, combining skin, organs, and bones.
• Probabilistic PCA was used in a pose-normalized space to model statistical variations while preserving joint kinematics.
• Model performance was assessed on registered and publicly available datasets.

Main Results:

• The model achieved an average error of 3.6 mm for bones and 8.8 mm for organs on a registered dataset.
• Using skin surface data or patient metadata resulted in combined bone-organ errors of 8.68 mm and 8.11 mm, respectively.
• Errors on the TotalSegmentator dataset were 5.10 mm for bones and 8.72 mm for organs.

Conclusions:

• Anatomically parameterized statistical shape models can be created accurately and efficiently.
• The proposed holistic model offers significant benefits for automation in various medical applications.
• The model enables the construction of shape models directly integrable into diverse medical applications.