Registration and fusion of retinal images-an evaluation study

Abstract

We present the results of a study on the application of registration and pixel-level fusion techniques to retinal images. The images are of different modalities (color, fluorescein angiogram), different resolutions, and taken at different times (from a few minutes during an angiography examination to several years between two examinations). We propose a new registration method based on global point mapping with blood vessel bifurcations as control points and a search for control point matches that uses local structural information of the retinal network. Three transformation types (similarity, affine, and second-order polynomial) are evaluated on each image pair. Fourteen pixel-level fusion techniques have been tested and classified according to their qualitative and quantitative performance. Four quantitative fusion performance criteria are used to evaluate the gain obtained with the grayscale fusion.

Automatic Correspondence on Medical Images: A Comparative Study of Four Methods for Allocating Corresponding Points

“…There are numerous methods for extracting points of interest from medical images. 8,[29][30] As the efficiency of those methods varies with the nature of the subject images, two techniques were employed throughout this study to obtain the initial control points:…”

“…(a) The method described in 29 which is optimized for detecting bifurcations in retinal images (b) The general purpose method adopted by Likar and Pernus 8 for obtaining points around edges and ridges on dental images, which is suitable for working with a large number of points In the first case, the number of extracted bifurcation points is determined automatically by the algorithm, while for the dental images, a fixed number of 200 control points were extracted from each reference image. …”

“…As mentioned, the adopted comparison procedure assumes a predefined set of control points on the reference image in order to estimate their correspondences. In case of sets I-IV, those points were allocated using two automatic point extraction algorithms: one for extracting 200 points of interest from dental images 8 (sets III and IV) and a separate one for allocating bifurcation points on the retinal data 29 (sets I and II). In the latter case, the number of extracted points depended on the morphology of the retinal images (number of bifurcations).…”

“…In this case, points were extracted in both images using the methodology described in. 29 Therefore, the methods illustrated before can be used to assess the four point correspondence methods both qualitatively, through visual assessment, and quantitatively, using the feature of merit quoted in Eq. 7.…”

A quadrature filter approach for registration accuracy assessment of fundus images

“…The most common way to quantify the registration accuracy of such images is to use the vessels' skeleton in the two images and evaluate the distance between them [3]. However, extracting the skeleton is not a trivial task.…”