Learning accurate rigid registration for longitudinal brain MRI from synthetic data

2025-01-22

AI Summary

Conventional rigid registration methods for longitudinal brain MRI suffer from insufficient accuracy—particularly in estimating subtle inter-scan translations and rotations—due to intra-subject anatomical variability and multi-contrast acquisition differences. Method: We propose the first deep learning framework specifically optimized for longitudinal rigid registration. Our approach features an anatomy-aware, scan-agnostic network architecture and introduces a novel intra-subject synthetic data augmentation strategy that jointly models rigid and fine-grained non-rigid transformations. This enhances generalization across anatomical variations and multi-contrast sequences (T1, T2, FLAIR). Results: Experiments demonstrate a 32% reduction in rigid transformation error compared to cross-subject baselines. The method achieves sub-pixel registration accuracy and robustness across contrasts, significantly improving reliability for clinical follow-up and quantitative disease progression analysis.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Rigid registration aims to determine the translations and rotations necessary to align features in a pair of images. While recent machine learning methods have become state-of-the-art for linear and deformable registration across subjects, they have demonstrated limitations when applied to longitudinal (within-subject) registration, where achieving precise alignment is critical. Building on an existing framework for anatomy-aware, acquisition-agnostic affine registration, we propose a model optimized for longitudinal, rigid brain registration. By training the model with synthetic within-subject pairs augmented with rigid and subtle nonlinear transforms, the model estimates more accurate rigid transforms than previous cross-subject networks and performs robustly on longitudinal registration pairs within and across magnetic resonance imaging (MRI) contrasts.

Problem

• Brain Image Registration
• Rigid Alignment
• Machine Learning Models

Innovation

• Optimization Model
• MRI Alignment
• Deformation Handling