Linear and Deformable Image Registration with 3D Convolutional Neural Networks

• DOI: 10.48550/arxiv.1809.06226
• Publication Date: 2018-01-01
• Authors: Stergios Christod...; Mihir Sahasrabudhe; Maria Vakalopoulou; Guillaume Chassagnon; Marie-Pierre Revel; Stavroula Mougiak...; Nikos Paragios
• Keywords: Image registration; Similarity (geometry)

Image registration and in particular deformable methods are pillars of medical imaging. Inspired by the recent advances deep learning, we propose this paper, a novel convolutional neural network architecture that couples linear within unified endowed with near real-time performance. Our framework is modular respect to global transformation component, as well similarity function while it guarantees smooth displacement fields. We evaluate performance our on challenging problem MRI lung registration, demonstrate superior state art elastic methods. The proposed deformation (between inspiration & expiration) was considered clinically relevant task interstitial disease (ILD) classification showed promising results.