New AM-FM analysis methods for retinal image characterization

We develop new amplitude-modulated frequency-modulated (AM-FM) based methods to address some issues associated with the semantic gap between visual and mathematical features presented by retinal diseases such as age-related macular degeneration (AMD). Through the processing of simulated and real, clinical retinal images we gain an understanding of the effects of basic morphological characteristics of lesions associated with AMD. Through synthetic simulations, we discuss how histograms of the instantaneous amplitude and the instantaneous frequency magnitude, extracted from different scales, can be used to differentiate between images of different sizes and edge sharpness, while maintaining invariance with respect to rotations. We show that AM-FM features extracted from low and very-low frequency scales can clearly differentiate between retinal images containing Temporal Concentrated Drusen (TCD) and Geographic Atrophy (GA). Shape, size, distribution and edge sharpness are visual features used by ophthalmologists in identifying lesions such as drusen. We propose the use of new AM-FM derived features to quantitatively define these visual descriptions.