Classified Scale-Invariant Feature Transform Feature Based Elastic Image Registration for 2-DE Gels
Qiaofeng Ou,Bangshu Xiong,Haodong Zhang,Yong Yang,Huisheng Zhang
DOI: https://doi.org/10.1166/jmihi.2015.1469
2015-01-01
Journal of Medical Imaging and Health Informatics
Abstract:Comparative analysis of 2-dimensional gel electrophoresis (2-DE) images aims at identifying changes of protein spot patterns between different samples. The comparison needs to be carried out automatically by computers due to the heavy workload and error-prone operation by hand. Image registration is a crucial step in the procedure. In this work, we propose an elastic registration scheme for 2-DE images based on classified scale-invariant feature transform (CSIFT) features. Firstly, the SIFT features are detected and matched with best bin first (BBF) algorithm. Then, using random sample consensus (RANSAC) and geometric similarity based match method, the features are classified into four classes: perspective transformed ones, elastically transformed ones, error-matched ones and unmatchable ones. Lastly, the former two classes of features are used as control points of thin-plate spline (TPS) transform. The proposed scheme is extensively tested on real 2-DE gel images from various experiments, and in comparison with spot-based methods, the proposed scheme exhibits better elastic registration, especially with higher automation degree and stability.