MeiBRD: Meta-Learning Intraoperative Biomechanical Residual Deformation

2026-06-15

AI Summary

Intraoperative liver registration is highly challenging due to large soft-tissue deformations and sparse observations. Traditional biomechanical models often introduce bias, while purely data-driven approaches suffer from poor generalization and physical implausibility. This work proposes a hybrid registration framework that leverages a linear biomechanical model as a prior and represents the residual deformation via a geometry-aware graph neural diffusion function defined on a 3D liver mesh. A feedforward meta-learner rapidly adapts this residual function using sparse correspondence points treated as context samples. By integrating biomechanical priors, geometry-aware attention mechanisms, and meta-learning, the method significantly outperforms existing baselines on phantom datasets, demonstrating superior accuracy and enhanced generalization—particularly under out-of-distribution geometries and deformation scenarios.

Abstract

Accurate intraoperative liver registration is challenging due to substantial soft-tissue deformation yet sparse intraoperative measurements. Biomechanical models regularize this ill-posedness with prior knowledge but exhibit persistent prediction bias due to simplifying assumptions, while data-driven learning solutions struggle with data efficiency, generalization, and physical plausibility. We propose a hybrid registration framework that adapts a biomechanical prior using sparse intraoperative correspondences. Rather than learning a full deformation field, we learn a residual deformation function that corrects linear biomechanical predictions, modeled as a graph neural diffusion function with geometry-aware attention over the 3D liver mesh. To enable long-range information transfer of sparse observations, we take a novel perspective of sparse intraoperative measurements as \textit{context} samples where input-output pairs of the residual deformation function are fully observed, casting the problem into learning-to-learn this residual function from intraoperative context samples with feedforward meta-learners. Experiments on a deformable liver phantom dataset demonstrate improved registration accuracy and generalization compared to rigid, biomechanical, and data-driven baselines, particularly for out-of-distribution geometries and deformations.

Problem

• intraoperative registration
• soft-tissue deformation
• biomechanical modeling
• data efficiency
• generalization

Innovation

• meta-learning
• residual deformation
• biomechanical registration
• graph neural diffusion
• sparse intraoperative measurements