Exploring shape changes in healthy bone growth through 3D spatiotemporal statistical shape models: A scoping review

Abstract

Objective: Analyzing population trends of bone shape variation can provide valuable insights into growth processes. This review aims to overview state-of-the-art spatiotemporal statistical shape modeling techniques, emphasizing their application to 3D skeletal structures during healthy growth.

Methods: We searched PubMed and Scopus for articles on statistical shape modeling using a pediatric spatiotemporal dataset of 3D healthy bone models. Dataset characteristics and details on the shape models' development, analyses, and potential clinical use were extracted.

Results: Fourteen studies were found eligible, modeling one or multiple lower limb bones, the mandible, the skull, and vertebrae. The majority applied Principal Component Analysis on point distribution models to create a statistical shape model. Shape variation was analyzed based on shape modes, representing a specific shape change as a part of the overall variance. Unscaled models resulted in a more compact statistical shape model than scaled models. The latter represented more subtle shape variations due to the absence of size differences between the bone models. Four studies reported a significant correlation between the first shape mode and age, indicating a relationship between that type of shape variation and growth. Three studies reconstructed 3D models using prediction features of statistical shape modeling. Measuring difference between predicted and actual anatomy resulted in Root Mean Squared Errors below 3 mm.

Conclusion: Spatiotemporal statistical shape modeling provides insight into modes of shape variation during growth. Such a model can be used to find predictive factors, like age or sex, and deploy these characteristics to predict someone's bone geometry.

Keywords: Bone growth; Morphological model; Patient-specific model; Statistical shape model; Three-dimensional imaging.

Fig. 1

The first two shape modes for a radius shape model. Each row shows the variation within a principal component between -3σ and + 3σ. In this example, PC1 captures a difference in overall size. PC2 represents variation in angulation in the coronal plane.

Fig. 2

Visualization of the development of an SSM, with the radius as example. Shape correspondence among the dataset is established using template fitting of the Point Distribution Models (top). After alignment, PCA is applied to generate the SSM (bottom).

Fig. 3

Visualization of the effect of scaling using synthetic data. Shape variation represented by the first shape at the left, and compactness plots at the right for (A) unscaled models and (B) scaled models.