SIRE: Scale-invariant, rotation-equivariant estimation of artery orientations using graph neural networks

Dieuwertje Alblas, Julian Suk, Christoph Brune, Kak Khee Yeung, Jelmer M. Wolterink

Abstract: Highlights•We present SIRE, a modular local estimator for local vessel orientation.•SIRE is a data-efficient model with potential for advanced vascular analysis.•SIRE uses graph neural network on spherical inputs to preserve rotational symmetry.•SIRE uses multi-scale weight sharing and parallel processing for scale invariance.•Symmetry preservations let SIRE generalize to vessels of any size and tortuosity.