Advances in skin cancer image analysis

Invasive and in situ malignant melanoma together comprise ne of the most rapidly increasing cancers in the world. Invaive melanoma alone has an estimated incidence of 68,720 and n estimated total of 8650 deaths in the United States in 2010. arly diagnosis is particularly important since melanoma can be ured with a simple excision if detected early. In the past, the priary form of diagnosis for melanoma has been unaided clinical xamination. In recent years, dermoscopy has provided increased bility to observe pigmented lesion structures. However, deroscopy remains difficult to learn and subjective, and several tudies have confirmed the limits of unaided dermoscopy. Newer echnologies, including digital dermoscopy, infrared imaging, mulispectral imaging, and confocal microscopy, have recently come o the forefront in providing greater diagnostic accuracy. These maging technologies can serve as an adjunct to physicians and n some implementations could provide fully automated skin caner screening potential. Although computerized techniques cannot s yet provide a definitive diagnosis, they can be used to improve iopsy decision-making as well as early melanoma detection in the linic, especially for patients with multiple atypical nevi, congenital evi, and other clinical presentations. The goals of this special issue are to summarize the state-ofhe-art in the computerized analysis of skin cancer images and to rovide future directions for this exciting subfield of medical image nalysis. The intended audience includes researchers and practicng clinicians, who are increasingly using digital analytic tools. The special issue opens with “Reconstruction of hyperspectral utaneous data from an artificial neural network-based multispecral imaging system” by Jolivot et al. The authors describe a novel ultispectral imaging system that includes a CCD camera, a rotatng wheel bearing a set of seven interference filters, a light source, nd a computer. The system employs a software solution to recontruct hyperspectral cubes from multispectral images by means f a heteroassociative neural network. The resulting cubes comine optical reflectance spectral data with two-dimensional spatial nformation. The special issue continues with two articles on enhancement f dermoscopy images. In “Automated color calibration method or dermoscopy images,” Iyatomi et al. describe a color correction ethod for dermoscopy images based on the Hue-Saturation-Value HSV) color model. First, a multiple linear regression model for each f the H, S, and V channels is built using various low-level features xtracted from a training image set. Using these regression models, he method then automatically adjusts the hue and saturation of a reviously unseen image. The advantage of such a supervised learnng approach is that it does not rely on any hardware or external