Purchase this article with an account.
A. M. Broehan, T. Rudolph, C. Tappeiner, S. P. Rothenbuehler, C. A. Amstutz, S. Wolf, J. H. Kowal; Validation of a Registration Algorithm for the Multimodal Spatial Alignment Between an Initial Retina Video Frame and a Retina Composite Image. Invest. Ophthalmol. Vis. Sci. 2010;51(13):1807.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Accurate placement of lesions is crucial for the effectiveness and safety of a retinal laser photocoagulation treatment. Computer assistance provides the capability for improvements to treatment accuracy and execution time. One possible solution for compensating retinal motion is to register the frames of a live retina video stream on a retina composite image potentially containing a treatment plan. We developed and validated a method for the multimodal registration of the initial frame of a video sequence on a retina composite image.
The general concept of the multimodal registration procedure is the following. Points on the vessel centerline and the optic disc are detected in both modalities. In a first step the initial video frame and the composite image are spatially pre-aligned using the determined optic disc parameter. Subsequently, the detected points on the vasculature from both images are iteratively matched starting in a small region around the optic disc and then expanding the matching region until the entire video frame is registered. The transformation model changes from affine to quadratic. For the optic disc detection we introduced a new approach, which calculates the highest local mean intensity variance at detected vessel centerline points to set up a region of interest for applying the Hough transform. The presented registration procedure is designed for a computer assisted retinal laser photocoagulation system. It initializes our real-time registration procedure that makes use of the spatial alignment information from the registered initial frame and the features of the segmented composite image to register subsequent frames of the video sequence.
The presented registration procedure was applied to 35 pairs of composite images and initial video frames acquired with a scanning digital ophthalmoscope (SDO). The recorded data cover a wide range of image quality and reveal diverse retinal pathologies. For 31 of the 35 registered composite image/SDO video frame pairs the vasculature of both images appeared well aligned when presented in checkerboard overlay and were hence considered successfully registered. The average registration error obtained was 2.68 pixels.
The algorithm demonstrated its capability to register various pairs of initial SDO video frames and composite images acquired from patients with respectable accuracy.
This PDF is available to Subscribers Only