Landmark-free Statistical Shape Modeling via Neural Flow Deformations

TL;DR

FlowSSM introduces a neural flow-based shape modeling method that captures anatomical variability without dense correspondence, outperforming existing models in robustness and expressiveness for medical shape analysis.

Contribution

It presents a novel neural flow-based approach for shape modeling that eliminates the need for dense correspondence, enhancing robustness and applicability.

Findings

Outperforms state-of-the-art shape priors for femur and liver.

Effectively separates healthy and pathological shapes.

Demonstrates accurate shape reconstruction from partial data.

Abstract

Statistical shape modeling aims at capturing shape variations of an anatomical structure that occur within a given population. Shape models are employed in many tasks, such as shape reconstruction and image segmentation, but also shape generation and classification. Existing shape priors either require dense correspondence between training examples or lack robustness and topological guarantees. We present FlowSSM, a novel shape modeling approach that learns shape variability without requiring dense correspondence between training instances. It relies on a hierarchy of continuous deformation flows, which are parametrized by a neural network. Our model outperforms state-of-the-art methods in providing an expressive and robust shape prior for distal femur and liver. We show that the emerging latent representation is discriminative by separating healthy from pathological shapes. Ultimately, we demonstrate its effectiveness on two shape reconstruction tasks from partial data. Our source code is publicly available (https://github.com/davecasp/flowssm).