A New Image Similarity Metric for Improving Deformation Consistency in Graph Based Groupwise Image Registration

Zhenyu Tang, Pew Thian Yap, Dinggang Shen

Research output: Contribution to journalArticle

Abstract

Graph-based groupwise image registration (G-GIR) aims to register a group of input images accurately without bias. In G-GIR, an image similarity metric (ISM) is used to construct a graph that links similar images with graph edges. From the built graph, a group center image and the shortest paths linking it to all other images can be determined. The deformation field aligning each image to the group center image can be obtained by composing sub-deformation fields which come from registration of adjacent images along the corresponding shortest path. The majority of ISMs used in G-GIR are based on image intensity. Since image intensity can be ambiguous and is not directly related to deformation directions, inconsistency problem in the sub-deformation fields along the shortest paths can occur. The inconsistency mentioned here refers to the directions of deformation vectors in the sub-deformation fields along each shortest path are significantly different or even opposite at corresponding positions. Such problem can make G-GIR inefficient and easily to be trapped in local minimum. In this paper, we propose a new ISM for G-GIR, by which consistency in the sub-deformation fields along the shortest paths can be significantly improved. We evaluate our method in comparison with three state-of-the-art ISMs using a common G-GIR framework. The experimental results with both toy and real images show that our method significantly improves registration accuracy.

Original languageEnglish
JournalIEEE Transactions on Biomedical Engineering
DOIs
Publication statusAccepted/In press - 2018 Jan 1

    Fingerprint

Keywords

  • Graph
  • groupwise image registration
  • image similarity metric
  • MR brain images

ASJC Scopus subject areas

  • Biomedical Engineering

Cite this