ART-based image analysis for pigmented lesions of the skin

Presents a robust, reliable computer-aided diagnostic tool for analyzing pigmented lesions of the skin, particularly malignant melanoma. The goal is to produce quantitative information to assist clinicians and researchers in diagnosing, monitoring and understanding the physiological processes of melanocytic lesions. The system described combines an adaptive resonance theory (ART) neural network with a comprehensive user interface and image analysis tools to extract quantitative information from color photographs of skin lesions. The ART operates on an RGB image, clustering the image into homogeneous regions. Connected component analysis extracts these regions and computes shape parameters and a color variegation metric. ART offers the advantages of well-understood theoretical properties, an efficient implementation, and clustering properties that are consistent with human perception