Statistical modeling and classification of reflectance confocal microscopy images

This paper deals with the characterization and classification of reflectance confocal microscopy images of human skin. The aim is to identify and characterize the lentigo, a phenomenon that originates at the dermo-epidermic junction of the skin. High resolution confocal images are acquired at different skin depths and are analyzed for each depth. Histograms of pixel intensities associated with a given depth are determined, showing that the generalized gamma distribution (GGD) is a good statistical model for confocal images. A GGD is parameterized by translation, scale and shape parameters. These parameters are estimated using a new estimation method based on a natural gradient descent showing fast convergence properties with respect to state-of-the-art estimation methods. The resulting parameter estimates can be used to classify clinical images of healthy and lentigo patients. The obtained results show that the scale and shape parameters are good features to identify and characterize the presence of lentigo in skin tissues.