Purchase this article with an account.
Reza Jafari, Charles A. Reisman; Vendor neutral multimodal registration for en face OCT and fundus images. Invest. Ophthalmol. Vis. Sci. 2019;60(11):PB0178.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To present an automated, robust, and accurate method to align images from different modalities (en face optical coherence tomography (OCT) versus color fundus images and color fundus versus FAF/FA/red-free images) using a novel multimodal registration technique.
OCT images and fundus photos of varying types are widely used in the diagnosis of eye diseases, and an automated registration routine helps clinicians to diagnose and monitor diseases from both imaging modalities. The proposed multimodal registration method is KAZE feature-based. After removing the background and enhancing contrast as a preprocessing for input images, KAZE features are detected and extracted. Inlier features are identified by calculating distances between corresponding features. The transformation matrix is then calculated to register the target and reference images. The proposed method was tested on 69 pairs of images captured by TRC-50DX and DRI OCT Triton (both Topcon Corp., Tokyo, Japan); images from other vendors were also used. The size of the OCT en face (macula or disc region, or both) and the fundus images (color, FAF, FA, red-free) are varied. The tested images were as follows: color to color, en face to color, FAF or FA to color, and red-free to color. The registration accuracy was evaluated using root mean square error (RMSE), which measured the degree of misalignment between feature points of reference images and corresponding feature points in target images.
By visual check, the registrations for all input image pairs were successful. Quantitatively, the mean accuracy was 1.35±0.52 pixel and the algorithm also performed well in the presence of artifacts, such as vignetting and media opacity.
A multimodal registration based on KAZE features was proposed and implemented for OCT en face, color, FAF, and red-free images. The method was tested on different scan modes and different resolutions. The experimental results suggest the algorithm is robust and accurate.
This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.
This PDF is available to Subscribers Only