Abstract
Color measurements (L*a*b* measures), standardized photographs or scoring scales are widely used to study rosacea. However, quantification methods involving the direct processing of images are not currently available, in particular with images of the microcirculatory network provided by videocapillaroscopy. This article describes a new image analysis method that is able to learn and quantify the extent and intensity of the rosacea. Images are taken of the cheekbone using a videocapillarospope with a ×50 magnification, which corresponds to an investigation area of approximately 27 mm2. The images obtained by this technique show mottling of more or less intense redness. Our system is based on an interactive approach where the investigator selects the zones affected by rosacea on the image (mottling), as well as the zones without signal (nonrosacea artefacts, e.g., white background). A neural filtering (linear perceptron) classifies these selections and generalizes them to other images. An automatic thresholding is applied to the neural restitution and permits us to compute the total surface of rosacea, as well as the number of distinct areas of rosacea. The neural filtering process detects the limits of the rosacea areas precisely and avoids the light areas without signal. The algorithm is implemented in software (Capilab Toolbox®) that allows us to carry out multicentric studies using various processing systems with different colorimetric characteristics.