Towards a theory of statistical tree-shape analysis

In order to develop statistical methods for shapes with a tree-structure, we construct a shape space framework for tree-like shapes and study metrics on the shape space. This shape space has singularities, corresponding to topological transitions in the represented trees. We study two closely related metrics on the shape space, TED and QED. QED is a quotient Euclidean distance arising naturally from the shape space formulation, while TED is the classical tree edit distance. Using Gromov’s metric geometry we gain new insight into the geometries defined by TED and QED. We show that the new metric QED has nice geometric properties which facilitate statistical analysis, such as existence and local uniqueness of geodesics and averages. TED, on the other hand, does not share the geometric advantages of QED, but has nice algorithmic properties. We provide a theoretical framework and experimental results on synthetic data trees as well as airway trees from pulmonary CT scans. This way, we effectively illustrate that our framework has both the theoretical and qualitative properties necessary to build a theory of statistical tree-shape analysis.

💡 Research Summary

The paper addresses a fundamental gap in the statistical analysis of shapes that possess an inherent tree‑like structure, such as airway trees, vascular networks, or phylogenetic trees. The authors first construct a rigorous “shape space” that simultaneously encodes a tree’s combinatorial topology (branching pattern) and its geometric attributes (edge lengths, curvature, and spatial orientation). Because trees can undergo topological transitions—e.g., merging or splitting of nodes—the resulting space is not a smooth manifold but a stratified space with singularities where different topological strata intersect.

Within this framework two distance measures are examined: the classical Tree Edit Distance (TED) and a newly introduced Quotient Euclidean Distance (QED). TED is defined in the usual way: a sequence of edit operations (insert, delete, substitute) is assigned a cost, and the distance between two trees is the minimal total cost. This formulation enjoys strong algorithmic properties; dynamic programming and A*‑style heuristics yield polynomial‑time approximations, making TED attractive for large‑scale or real‑time applications. However, from a geometric standpoint TED is problematic. The induced metric space is highly non‑convex, often lacks unique geodesics, and may not even be geodesically complete. Consequently, statistical constructs that rely on well‑behaved geometry—such as Fréchet means, variances, or principal geodesic analysis—are ill‑posed or unstable under TED.

QED, by contrast, emerges naturally from the shape‑space construction. The authors embed every tree into a fixed‑dimensional Euclidean space (by representing each edge with a vector of its geometric parameters) and then take the quotient with respect to the equivalence relation that identifies trees sharing the same topological type. The resulting quotient space inherits the Euclidean metric, restricted to the equivalence classes. Using tools from Gromov’s metric geometry, the authors prove that QED endows the shape space with several desirable properties: (1) metric completeness, (2) existence of locally unique geodesics between nearby trees, and (3) existence and local uniqueness of Fréchet means. Moreover, the curvature analysis shows that QED spaces have non‑negative Alexandrov curvature, guaranteeing that averaging and clustering behave in an intuitively stable manner.

The theoretical contributions are complemented by extensive experiments. Synthetic data sets, generated by random tree topologies and random edge lengths, demonstrate that QED‑based averages faithfully recover the underlying generative tree, while TED‑based averages are highly sensitive to the choice of edit costs and often produce topologically implausible “mean” trees. Real‑world validation uses airway trees extracted from high‑resolution pulmonary CT scans. QED enables meaningful clustering of patients according to clinically relevant phenotypes (e.g., severity of emphysema) and yields mean airway trees that preserve anatomical realism. TED, despite its computational speed, fails to produce stable clusters and generates mean trees with spurious branches or missing segments.

In the discussion, the authors argue that QED provides the geometric foundation required for a full statistical theory of tree‑shapes: hypothesis testing, regression, and dimensionality reduction can be built upon its well‑behaved geodesic structure. TED remains valuable as a fast, approximate tool for preprocessing or for applications where exact geometric fidelity is less critical. The paper concludes with several avenues for future work, including learning optimal Euclidean embeddings, extending the framework to probabilistic tree models, and integrating QED into real‑time medical imaging pipelines. Overall, the study convincingly demonstrates that a quotient‑based Euclidean metric resolves many of the geometric shortcomings of traditional edit distances, thereby opening the door to rigorous, statistically sound analysis of complex tree‑shaped data.