Abstract—Motivated by the problem of retinal image registration, this paper introduces and analyzes a new registration algorithm called Dual-Bootstrap Iterative Closest Point (Dual-Bootstrap ICP). The approach is to start from one or more initial, low-order estimates that are only accurate in small image regions, called bootstrap regions. In each bootstrap region, the algorithm iteratively: 1) refines the transformation estimate using constraints only from within the bootstrap region; 2) expands the bootstrap region; and 3) tests to see if a higher order transformation model can be used, stopping when the region expands to cover the overlap between images. Steps 1): and 3), the bootstrap steps, are governed by the covariance matrix of the estimated transformation. Estimation refinement [Step 2)] uses a novel robust version of the ICP algorithm. In registering retinal image pairs, Dual-Bootstrap ICP is initialized by automatically matching individual vascular landmarks, and it aligns images based on detected blood vessel centerlines. The resulting quadratic transformations are accurate to less than a pixel. On tests involving approximately 6000 image pairs, it successfully registered 99.5 % of the pairs containing at least one common landmark, and 100 % of the pairs containing at least one common landmark and at least 35 % image overlap. Index Terms—Iterative closest point, medical imaging, registration, retinal imaging, robust estimation.

This paper describes an algorithm to continually and accurately estimate the absolute location of a diagnostic or surgical tool (such as a laser) pointed at the human retina, from a series of image frames. We treat the problem as a registration problem, using diagnostic images to build a spatial map of the retina and then registering each on-line against this map. Since the image location where the laser strikes the retina is easily found, this registration determines the position of the laser in the global coordinate system defined by the spatial map. For each on-line image, the algorithm computes similarity invariants, locally valid despite the curved nature of the retina, from constellations of vascular landmarks. These are detected using a high-speed algorithm that iteratively traces the blood vessel structure. Invariant indexing establishes initial correspondences between landmarks from the on-line image and landmarks stored in the spatial map. Robust alignment and verification steps extend the similarity transformation computed from these initial correspondences to a global, high-order transformation. In initial experimentation, the method has achieved 100% success on 1024 1024 retina images. With a version of the tracing algorithm optimized for speed on 512 512 images, the computation time is only 51 milliseconds per image on a 900MHz Pentium III processor and a 97% success rate is achieved. The median registration error in either case is about 1 pixel.

Real-time spatial referencing is an important alternative to tracking for designing spatially-aware ophthalmic instrumentation for procedures such as laser photocoagulation and perimetry. It requires independent, fast registration of each image frame from a digital video stream (1024 1024 pixels) to a spatial map of the retina. Recently, we have introduced a spatial referencing algorithm that works in three primary steps: (1) tracing the retinal vasculature to extract image feature (landmarks), (2) invariant indexing to generate hypothesized landmark correspondences and initial transformations, and (3) alignment and veri cation steps to robustly estimate a 12-parameter quadratic spatial transformation between the image frame and the map. The goal of this paper is to introduce techniques to minimize the amount of computation for successful spatial referencing. The fundamental driving idea is to make feature extraction subservient to registration and therefore only produce the information needed for veri ed, accurate transformations. To this end, the image is analyzed along one-dimensional, vertical and horizontal grid lines to produce a regular sampling of the vasculature, needed for step (3) and to initiate step (1). Tracing of the vascular is then prioritized hierarchically to quickly extract landmarks and groups (constellations) of landmarks for indexing. Finally, the tracing and spatial referencing computations are integrated so that landmark constellations found by tracing are tested immediately. The resulting implementation is an order-of-magnitude faster with the same success rate. The average total computation time is 45.5 milliseconds per image on a 2.2 GHz Pentium Xeon processor.

Motivated by the goals of improving detection of low-contrast and narrow vessels and eliminating false detections at non-vascular structures, a new technique is presented for extracting vessels in retinal images. The core of the technique is a new likelihood ratio test that combines matched filter responses, confidence measures and vessel boundary measures. Matched filter responses are derived in scale-space to extract vessels of widely varying widths. A vessel confidence measure is defined as a projection of a vector formed from a normalized pixel neighborhood onto a normalized ideal vessel profile. Vessel boundary measures and associated confidences are computed at potential vessel boundaries. Combined, these responses form a 6-dimensional measurement vector at each pixel. A learning technique is applied to map this vector to a likelihood ratio that measures the "vesselness" at each pixel. Results comparing this vesselness measure to matched filters alone and to measures based on the intensity Hessian show substantial improvements both qualitatively and quantitatively. When the Hessian is used in place of the matched filter, similar but less-substantial improvements are obtained. Finally, the new vesselness likelihood ratio is embedded into a vessel tracing framework, resulting in an e#cient and e#ective vessel extraction algorithm.

This paper presents a disease-oriented evaluation of two recent retinal image registration algorithms, one for aligning pairs of retinal images and one for simultaneously aligning all images in a set. Medical conditions studied include diabetic retinopathy, vein occlusion, and both dry and wet age-related macular degeneration. The multi-image alignment worked virtually flawlessly, missing only 2 of 855 images. Pairwise registration, the Dual-Bootstrap ICP algorithm, worked nearly as well, successfully aligning 99.5% of the image pairs having a su#cient set of common features and 78.5% overall. Images of retinas having an edema and pairs of images taken before and after laser treatment proved the most di#cult to register.