Purchase this article with an account.
A. M. Broehan, J. H. Kowal; Online Registration of Retina Video Streams on a Planning Modality. Invest. Ophthalmol. Vis. Sci. 2009;50(13):305.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Placing laser lesions accurately is essential for an effective and safe laser photocoagulation treatment. Computer assistance during laser photocoagulation provides improvements with respect to accuracy and patient safety. To account for patients’ eye movements during treatment execution a successful computer assisted system has to reliably track the retina. We present a method for the online registration of retina video sequences acquired with a Scanning Digital Ophthalmoscope (SDO) on a retina laser treatment plan e.g. a retina composite image.
We have developed a new approach for the online registration of retina video sequences. In a first step the vessel centerline of the planning modality is extracted using an iterative vessel tracing procedure. The first video frame is then registered against the planning modality using our feature-based multi-modal retina image registration technique. Further on, each video frame is registered on the planning modality by first extracting the translational motion between the precedent and the current frame and then using the translational component combined with the registration matrix of the precedent frame to initialize a vessel point matching procedure in order to get the registration matrix for the current frame. The translational motion between two successive frames is obtained using normalized cross correlation. For the vessel point matching procedure, points on the vasculature of a retina video frame are extracted using opposite signed gradient values along horizontal and vertical grid lines. The final registration matrix is calculated by minimizing the distances between the extracted vessel points on the video frame and the vessel centerline of the planning modality by using the robust error norm of an M-estimator and iterative re-weighted least squares. The algorithm is implemented in the Medical Application Framework Marvin.
Our method was applied to a variety of SDO video sequences and retina composite images acquired from patients. Several hundred video frames could be registered with good accuracy as observed by visual inspection. In cases where the registration fails due to e.g. motion blur and eye closure, the registration procedure has to be reinitialized.
The new strategy shows very promising results to automatically register retina video frames on a planning modality which is essential for a computer assisted retina laser photocoagulation system to ensure a match between a previously established treatment plan and the current retina position.
This PDF is available to Subscribers Only