Abstract
Purpose :
To present an automated method for robust and accurate multimodal registration of optical coherence tomography (OCT) en face images and corresponding color fundus images.
Methods :
The proposed multimodal registration framework is based on KAZE features. After a preprocessing stage for input images, KAZE features are extracted. The best matching features are identified by calculating the distance between their corresponding features. A transformation matrix is then applied to register the color fundus and en face images. The proposed method was tested on 48 pairs of OCT en face images and corresponding color fundus images captured by DRI OCT Triton (Topcon Corp. Tokyo, Japan). The size of OCT en face images varied depending on OCT scan mode, covering the macula or disc region, or both. The registration accuracy was evaluated using root mean square error (RMSE), which measures the amount of misalignment between feature points of en face images and corresponding feature points in color fundus images.
Results :
The registrations for all input image/photo pairs were successful by visual check. Quantitatively, the mean accuracy was 31.85±10.98 µm, and the algorithm also performed well in the presence of artifacts, such as vignetting and media opacity. An example of the proposed registration is shown in Figure 1.
Conclusions :
A multimodal registration for aligning OCT en face and color fundus images based on KAZE features was proposed and implemented. The proposed method was tested on different scan modes, and experimental results show that it is robust and accurate.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.