Abstract
Purpose: :
Accurate placement of lesions is crucial for the effectiveness and safety of a laser photocoagulation treatment. Computer assistance has the capability for 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. The aim of the presented study is to investigate different published algorithms concerning their ability to solve the real-time registration problem during computer assisted laser photocoagulation and to evaluate different retina imaging modalities regarding their potential to be used for intra-operative image acquisition.
Methods: :
Within the study image and video data from both eyes of 35 patients were recorded. Central fundus images were acquired using a non-mydriatic fundus camera VISUCAM NM/FA (Carl Zeiss Meditec, Jena, Germany). Video data streams were captured with a laser slit lamp LSL 532s with laser VISULAS 532s (Carl Zeiss Meditec, Jena, Germany) equipped with a DigiCam beam splitter adapter and a scanning digital ophthalmoscope SDO-A (WILD Medtec, Wien, Austria). The fundus images served as planning modality against which the intra-operatively acquired video streams were registered. To investigate retina tracking methods the concept was to initially co-register the planning modality and the first retina video frame using a feature-based registration algorithm that iteratively traces the vasculature. The obtained transformation parameters were then updated by tracking the frame-to-frame motion. After reviewing several approaches for real-time co-registration two strategies were selected: an intensity-based template matching using different similarity metrics and a method based on the construction of a two-dimensional retinal vessel template.
Results: :
Applied to the acquired patient data the presented algorithms revealed high potential for solving the real-time registration task, whereas the feature-based method seemed to be more accurate and robust. The ratio between successful frame-to-frame registrations and tracking-loss events was significantly higher for SDO video streams compared to slit lamp video streams. Additionally, the employment of red-free filters improved the tracking reliability considerably for both devices.
Conclusions: :
The feature-based method would most probably be the preferred tracking strategy. As an intra-operative imaging device the SDO appeared to be more appropriate compared to the slit lamp due to the larger field of view and better achievable image quality.
Keywords: image processing • retina • laser