The Case for Interpretable Geometry: Statistical Shape Models vs. Curvature-Based Descriptors in Aortic Disease Classification
Year: 2026
Type: article
Abstract: Purpose Quantifying aortic morphology is central to surgical planning for thoracic endovascular aortic repair (TEVAR), yet no consensus exists on how best to represent three-dimensional aortic shape for outcome prediction. Two broad strategies have emerged: statistical shape analysis (SSA), which relies on statistical methods and dimensionality reduction to capture the most significant shape modes, and geometrically-informed approaches that extract descriptors grounded in differential geometry. Here, we directly compare these paradigms on a cohort of 290 CTA scans classified by surgical outcome (non-pathological, successful TEVAR, failed TEVAR). Methods For the geometrically-informed approach, we use a two-dimensional feature space using fluctuation in integrated Gaussian curvature $$\delta K$$ δ K and mean aortic radius R . For SSA, we construct a point-cloud shape model with dimensionality reduction using Principal Component Analysis (PCA) and evaluate classification performance as a function of the number of retained principal components. Both approaches are evaluated by repeated cross-validation grouped by patient, with all scaling, PCA, and hyperparameter selection performed within the training folds, and by external testing on scans from the GORE TAG 08-01 trial. Results SSA’s leading principal components encode variations in global aortic size and shape and are statistically redundant with ( R , $$\delta K$$ δ K ), yet they lack a one-to-one correspondence with interpretable anatomical quantities. With the same number of dimensions, the geometric descriptors classified at least as well as SSA (Gaussian process classification, cross-validated weighted $$F_1$$ F 1 of 85.8% versus 84.9%). SSA’s cross-validated $$F_1$$ F 1 peaked with two principal components (83.1%) and declined as more were added, although its training-set accuracy rose above 90%, which indicates overfitting. On the independent trial cohort, in which every patient required TEVAR, SSA classified 50.7% of scans as non-pathological, compared with 21.2% for the geometric approach, suggesting better generalizability. Conclusion Two physically defined descriptors classified aortic disease at least as well as the leading SSA modes and transferred better to an independent cohort. SSA modes offer an intuitive visualization of how shape varies across a cohort, but geometric descriptors, which are measured identically for any patient and do not depend on the cohort used to build a model, provide a more robust basis for patient-level classification.
Authors: Duc Nguyen, Joseph A. Pugar, Luka Pocivavsek
Institution: University of Chicago
Language: English
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
Estimated APC paid: $3,790
OpenAlex ID: W7171972893
