Purpose
Fluorescein angiography in Age-related macular degeneration (AMD) represents the gold standard for classification into different clinical forms. In this context a good quality of images acquired during angiographic phases is required for reliable diagnosis. However, the quality of images is frequently impaired by media opacities. This study aims to develop a new tool on the Spectralis platform of Heidelberg Engineering to improve the quality of images.
Methods
A simplest Partial Derivative Equations (PDE) for smoothing an image is to apply a linear diffusion process that corresponds to the heat equation in the physical context. This iterative approach smooths and blurs structural features such as edges of objects in the image. To resolve this problem, Perona et al have proposed a non-linear and space- variant transformation named anisotropic diffusion to reduce the smoothing effect near edge and Gilboa et al have proposed the ramp preserving denoising that consists to generalize the linear scale spaces in the complex domain. To reduce the blur effect of confocal images, we propose a new numerical scheme based on PDE. This processing allows as well to smooth images with edge preservation for a selection and the geometrical study of AMD lesions. We adapted the principle of complex diffusion for several directions using directional PDEs with the concept of hypercomplex numbers to apply linear and nonlinear diffusion schemes for angiographic pictures. The proposed non-linear diffusion lead to a good smoothing of images with a Gaussian noise and Poisson noises.
Results
A Gaussian noise on a classical Shepp-Logan phantom image (Signal-to-noise ratio SNR=17,75dB) was added to allow a good restoration (SNR=32,7 dB) compared to the ramp preserving denoising (SNR=15,2dB) and to the Perona-Malick diffusion (SNR=32dB). Different angiographic pictures from eight AMD patients were improved using this new numerical scheme.The ramp preserving denoising does not preserve edges in the Poisson noise (average SNR=10,4 dB) unlike the two other non-linear diffusion. The Perona-Malick diffusion smoothes the homogeneous zones (average SNR=16,0dB) but not as well as the hypercomplex non linear diffusion (average SNR=17,1dB).
Conclusions
This study provides a new useful tool to improve the quality of angiographic pictures and thus their interpretation. This new tool could be useful in case of media opacities.
Keywords: 549 image processing •
550 imaging/image analysis: clinical