Session

Friday, June 23, 2017 - 09:00 to 18:45
Automatic Segmentation of Melanoma in Dermoscopy Images using Fuzzy Numbers
Abstract: 
Melanoma is the most dangerous type of skin cancer, but when treated in its early stages the chance of cure is increased. However, the detection of melanoma is a challenging task even for specialists due to low contrast of skin lesions and presence of artifacts. Therefore, developing an automatic segmentation tool for skin lesion analysis using dermoscopy images is a critical step for improving the diagnosis. This work proposes an automatic melanoma segmentation approach, based on Fuzzy Numbers. The technique was evaluated using 571 images from ISDI data set, composed of 446 benign and 125 malignant melanoma. The proposed approach was compared with three state-of-art techniques and was evaluated through the metrics of sensitivity, specificity, Jaccard index, and balanced accuracy. Results show that the fuzzy approach has a better segmentation compared to the other techniques, obtaining values of 0.77 of sensibility, 0.94 of specificity, 0.65 of the Jaccard index, and 0.85 of balanced accuracy. Results demonstrate that the segmentation approach using fuzzy num- bers is highly competitive with all algorithms analyzed.
Filipe Rolim Cordeiro's picture
Filipe Rolim Cordeiro
Federal Rural University of Pernambuco (BR)
Jessica Barbosa Diniz's picture
Jessica Barbosa Diniz