Uma metodologia híbrida para segmentação de lesões de pele

Abstract

Computer aided methods are widespread demanded in medical applications. As such, methodologies to automatic extract contours are desired to aid the automatic diagnosis of skin lesions. In this work, we propose a hybrid method to detect and extract skin lesion contours from dermatoscopic images. In the proposed method, to obtain the contour that includes all lesion regions, the region growing technique, based on a Quadtree implementation, is used to extract an initial contour. Afterwards, this contour is refined by using a traditional active contour model. Experimental results indicate that the proposed method is promising to detect skin lesion areas and to extract their contours from dermatoscopic images. Actually, the extracted contours maintain the original lesion features that are usually used in their diagnosis. Additionally, the results allow concluding that the method is able to detect lesion regions even in images with strong noise, like in images of the scalp.