Supervised larning of ANN for plaque lesion discrimination based on chromatic color indices

Abstract

This paper presents about recognizing selected skin lesions through automation diagnosis system as an aid for dermatologist. Previous work has shown that analyzing psoriasis lesion digital images using YCbCr color model is feasible. This is because human skin images if digitally quantified operates better on the chrominance plane rather than the normal RGB. The statistical outcomes from the previous work have given valuable information about discrimination of plaque from the other psoriasis lesions. This research will use the above information in the designing of an automation diagnosis model for plaque discrimination. YCbCr mean gradation color indices representing cropped images of psoriasis lesion are used to train 3 layer perceptron artificial neural networks (ANN). Backpropagation supervised learning of the layer networks are optimized by observing the performance indicators regularly applied in medical research. In this study the skin lesion classification is using YCbCr reflectance indices for psoriasis lesion that focused on guttate, plaque and erythroderma. In this work, components of YCbCr for differential method is applied to the input of ANN model with 400 samples used for training while 200 samples is for testing. © 2008 Springer-Verlag.