AtlasODE: Learning Continuous Atlases via Neural Ordinary Differential Equations

Jichang zhang, Feng li, Tongtong Che, Shuyu Li

Keywords: Anatomical atlas, Brain template, Neural ODE, Aging and disease evolution

TL;DR: AtlasODE reframes brain anatomical atlas generation as continuous-time integration, capturing the morphological manifold while providing an interpretable mechanism to adaptively model disease evolution.

Registration Requirement: Yes

Abstract: Current approaches to brain atlas generation rely on independent static fitting, inherently failing to capture continuous morphological evolution. To address this gap, we propose AtlasODE, a novel framework leveraging Neural Ordinary Differential Equations to reframe atlas generation as a continuous-time integration process. To explicitly model neurodegeneration, our Pathological Trajectory Modulator (PTM) formulates disease as an adaptive perturbation, characterizing pathological evolution as a divergence from the normative aging trajectory. Experimental results on synthetic toy data and brain MRI data demonstrate the superior performance of AtlasODE.

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 30