Abstract
Abstract: :
Purpose: To develop a precise and rapid algorithm for the automated registration of sequential fluorescein frames that allows for registration of multimodal images. Methods: Stock digital images with all the patient–identifiable features removed were culled from an electronic database. We sought patients who had undergone multimodal imaging, e.g. fluorescein angiography, indocyanine green angiography, fundus photography, red–free photography. Images were obtained with the same Zeiss fundus camera, both on the same day and at later intervals. An algorithm was developed using the Java coding language that rapidly compared corresponding pixel regions on sequential frames of the same eye, regardless of image modality. Briefly, a combination of random permutations and image analysis kernels are used to find the best consensus image between two sequential frames. Results:The robust nature of this algorithm allows it to handle up to 25 images in a stack, automatically align and resize these images, and allow for rapid progression either forward or backward through the stack. This algorithm is highly reproducible and has a low error rate. Computational algorithms also analyze the stacked images to identify the regions of statistically significant luminosity changes over time. A false color map of the retina captures temporal variation in fluorescence intensity for rapid diagnosis. Finally, in the case of unimodality imaging, a pseudomovie can be developed to show the dynamic filling and dye dilution with a corresponding pseudocolor movie off to the side. Conclusions: We have developed an accurate, reproducible, and efficient algorithm for automated registration of multimodal fundus images. Future modifications may result in a clinically useful tool for research and diagnostics.
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • imaging/image analysis: clinical • retina