Biomedical Signal Processing and Control 45 (2018) 325–338 Contents lists available at ScienceDirect Biomedical Signal Processing and Control jo ur nal homepage: www.elsevier.com/locate/bspc A-RANSAC: Adaptive random sample consensus method in multimodal retinal image registration Zahra Hossein-Nejad a , Mehdi Nasri b,∗ a Department of Electrical Engineering, Sirjan branch, Islamic Azad University, Sirjan, Iran b Young researchers and elite club, Khomeinishahr branch, Islamic Azad University, Khomeinishahr, Iran a r t i c l e i n f o Article history: Received 23 December 2017 Received in revised form 13 May 2018 Accepted 6 June 2018 Keywords: Angiographic imaging Image registration Retinal images SIFT RANSAC a b s t r a c t In this paper, an adaptive Random Sample Consensus (A-RANSAC) method is proposed for multimodal retinal image registration. In this method, the features of two various images from images taken with different modalities such as FA (Fluorescein angiography) and RF (Red free) are extracted using a modified version of Scale Invariant Feature Transform method (SIFT) called SAR-SIFT which is originally used for Synthetic Aperture Radar images. Then, the matching performance between these images is enhanced using the proposed A-RANSAC. In the A-RANSAC method, the threshold value is chosen so that the Root Mean Square Error (RMSE) and the number of removed matches are optimized simultaneously. The efficiency of the proposed method has been investigated in other modes such as high resolution and low-quality retinal image registration in addition to multimodal registration. The simulation results on several retinal image datasets show that the proposed method improves the precision matching by 9.89% and rate of success by 25% on the average compared to the SAR-SIFT method. © 2018 Elsevier Ltd. All rights reserved. 1. Introduction Fluorescein angiographic (FA) is done to diagnose retinal prob- lems and diseases related to it such as diabetic retinopathy, and age-related macular degeneration [1,2]. FA is taken using Scan- ning Laser Ophthalmoscope (SLO) that, one or two images before injection of sodium fluorescein dye and a few images at regular intervals after the injection [3,4]. Normally, fluorescein does not leak from retinal vessels (including late venous and arteriovenous) and remains in the vessel space. If blood vessel walls are abnormal, this dye may leak into the retina. In this case, vessels obstruction, damage to the lower layers of the retina, or appearance of new blood vessels with abnormal growth underneath may emerge. To identify and assess the progression of the disease, it is necessary to perform image registration between different phases (arterial phase, arteriovenous phase, venous phase, late venous phase, and recirculation phase) of FA, which ultimately can help ophthalmol- ogists to diagnose and provide better treatment planning [5,6]. Image registration is a key component in automatic image anal- ysis and this process has an important role in medical images including retina. Retinal image registration is the process of find- ing geometric transformations between two or more images of ∗ Corresponding author. E-mail addresses: hoseinnejad.zahra@yahoo.com (Z. Hossein-Nejad), nasri me@iaukhsh.ac.ir (M. Nasri). the retina, taken at different times, different viewpoints or by dif- ferent sensors [7,8]. For various reasons, due to reduction of the retina images’ quality, a general approach is not applicable to these images. However, registration methods of retina images in general can be divided into area-based and feature-based [9–11]. Area-based registration methods directly use gray (intensity) surface distribution of images in the windows with the same dimensions. For this purpose, use similarity metric, geometric transformation parameters between images are determined. Com- mon similarity metrics for retinal angiography images include mutual information (MI) [12,13], entropy correlation coefficient (ECC) [14,15], and phase correlation [16]. Performance of MI, when there are changes in the texture of the retinal image and changes in scale, is not suitable [17,18]. When there is high translation and content changes between images, phase-based methods do not have a good performance [19]. Thus, overall it can be con- cluded that area-based methods face different challenges. One of these challenges is that when illumination changes between the images are high, overlapping areas in the images are low [7,20]. In this case, when changes in the texture and content are high, this category of methods has unsuitable performance. In addition, the run-time of area-based methods is high, as they use entire content of the image [21]. In references [22,23], comparisons have been con- ducted between registration methods showing that feature-based methods have better performance for retina images registration. Feature-based registration methods detect salient features of images and then calculate matching and parameters of transforma- https://doi.org/10.1016/j.bspc.2018.06.002 1746-8094/© 2018 Elsevier Ltd. All rights reserved.