OCT registration using macular thickness map and curvature maps

Abstract

Purpose : Registration of longitudinal OCT data is important for measuring the changes of retinal thicknesses over time. The registration based on OCT en face images becomes difficult when the OCT data is generated by a low-cost OCT system due to low contrast in the data. Here we propose to use the macular thickness and principal curvatures as alternative maps to provide sufficient number of landmarks for registration.

Methods : To obtain well distributed landmarks across the OCT field of view of two maps for robust registration, the macular thickness and corresponding principal curvature maps were used, where landmark correspondences were extracted from all three maps of each scan. Then, a subset of landmark correspondences with high confidence were selected using an exhaustive search method to compute the rigid transformation. Figure 1 shows an example of the thickness map registration. The mean registration error along with the number of landmarks between the landmark matches after registration were calculated for each pair of OCT volumes.

Results : Figure 2 shows the statistics for the registration error and landmark distribution for 243 pairs of OCT volumes of disease eyes acquired using a low-cost OCT prototype system (ZEISS, Dublin, CA). The mean and standard deviation of registration error in x, y and radial direction xy (sqrt(x^2+y^2)) are smaller than 70 and 15 microns respectively, which is acceptable for macular thickness analysis. The average number of landmarks used for registration is relatively high for computing three parameters of rigid transformation.

Conclusions : A registration method based on macular thickness and curvature maps has been explored in this study. We showed that this algorithm performed well for data acquired from low-cost OCT devices.

This is a 2021 ARVO Annual Meeting abstract.

 

Figure 1 Registration examples of using thickness map only (a-d), max curvature map (e-h), min curvature map (i-l), and the final result (m) with all the landmarks (n) from these 3 individual methods (d, h, l). The 1st and 2nd column show the reference and unregistered thickness maps and their corresponding curvature maps. The 3rd and 4th columns are the registered images for each individual method (black lines illustrate the borders of the images in 2nd column) and the landmarks with resulting RMS error. The final registration (m) shows the most uniform distribution of the landmarks and lowest RMS error (n).

 

Figure 2 Statistics for the registration error