Representing Ambiguity in Registration Problems with Conditional\n Invertible Neural Networks

Le résumé fourni par la source

Image registration is the basis for many applications in the fields of\nmedical image computing and computer assisted interventions. One example is the\nregistration of 2D X-ray images with preoperative three-dimensional computed\ntomography (CT) images in intraoperative surgical guidance systems. Due to the\nhigh safety requirements in medical applications, estimating registration\nuncertainty is of a crucial importance in such a scenario. However, previously\nproposed methods, including classical iterative registration methods and deep\nlearning-based methods have one characteristic in common: They lack the\ncapacity to represent the fact that a registration problem may be inherently\nambiguous, meaning that multiple (substantially different) plausible solutions\nexist. To tackle this limitation, we explore the application of invertible\nneural networks (INN) as core component of a registration methodology. In the\nproposed framework, INNs enable going beyond point estimates as network output\nby representing the possible solutions to a registration problem by a\nprobability distribution that encodes different plausible solutions via\nmultiple modes. In a first feasibility study, we test the approach for a 2D 3D\nregistration setting by registering spinal CT volumes to X-ray images. To this\nend, we simulate the X-ray images taken by a C-Arm with multiple orientations\nusing the principle of digitially reconstructed radiographs (DRRs). Due to the\nsymmetry of human spine, there are potentially multiple substantially different\nposes of the C-Arm that can lead to similar projections. The hypothesis of this\nwork is that the proposed approach is able to identify multiple solutions in\nsuch ambiguous registration problems.\n

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.