Figure 1.

Summary of the methodology. Step 1: Extraction of raw images from each imaging modality at each visit. Note that images vary in field of view due to use of different imaging modalities. Step 2: Pre-processing of images as described in Ly et al.³³ Images are registered, background corrected, and contrast enhanced. Step 3.a: Clustering analysis using pattern recognition is performed across the three images yielding a single multispectral image where each color represents a distinct spectral theme class. Step 3.b Spectral theme classes assigned as drusen/not drusen using reference color fundus photograph annotated for drusen by clinical experts. Step 4: Identification of spectral theme classes attributed to drusen from the pseudocolored image. (A): Spectral theme classes identified as drusen were thresholded from all other theme classes in the pseudocolored image and (B) Drusen spectral theme classes were merged to create a duochrome image. Step 5: Quantification of drusen. Total number of pixels attributed to drusen were quantified from duochrome images at each visit and drusen change calculated as a percentage difference between the two visits. CFP: Color fund photograph, FAF: fundus autofluorescence, IR: infrared, OG: green scanning laser ophthalmoscopy images.