From Skin to Skeleton: A Statistical Shape Modelling Approach for Predicting Hand and Foot Bony Geometry

In clinics, musculoskeletal assessment mainly relies on conventional imaging which is expensive, involves radiation exposure, and lacks accessibility. Inferring skeletal morphology from 3D body surface scans shows potential as an alternative screening aid, although lacks detail in particular at extremities. This study leverages Statistical Shape Models (SSMs) to estimate position and orientation of the hand and foot bones from the skin surface.

Two datasets (140 feet, 79 hands with diverse morphologies and poses) were collected. For each dataset, a coupled skin-bone SSM was created. A nested cross-validation approach was used to optimize hyperparameters and prevent overfitting while fitting the isolated skin model of the coupled SSM to unseen skin data to infer bone structure and position.

For the feet, the mean absolute error (MAE) was 1.68 mm, with the highest errors occurring at the hindfoot. Similarly, for the hands, the MAE reached 1.37 mm, with the largest deviations observed at carpal bones.

This study demonstrates the feasibility of predicting bone morphology from skin surfaces using SSM-based shape completion, offering a potential non-invasive and accessible alternative to traditional imaging for musculoskeletal assessments.