Evaluation of Levenberg–Marquardt neural networks and stacked autoencoders clustering for skin lesion analysis, screening and follow‐up

• DOI: 10.1049/iet-cvi.2018.5195
• Publication Date: 2018-08-04T02:26:25Z
• Authors: Francesco Rundo; Sabrina Conoci; Giuseppe L. Banna; Alessandro Ortis; Filippo Stanco; Sebastiano Battiato
• Keywords: skin lesion dermoscopy images; skin lesion analysis; Computer applications to medicine. Medical informatics; stacked autoencoders; R858-859.7; dermoscopy image; 02 engineering and technology; visual analysis; cancer; image classification; skin; image segmentation; feature extraction; neural nets; biomedical optical imaging; medical image processing; stacked autoencoders; skin lesion analysis; visual analysis; morphological analysis; skin lesion dermoscopy images; dermoscopy image; SC-cellular neural networks; ad-hoc grey-level skin lesion image; ad-hoc clustering; benign against melanoma; hand-crafted image features; Levenberg-Marquardt neural network; Software; 1707; QA76.75-76.765; 0202 electrical engineering, electronic engineering, information engineering; morphological analysis; Computer software

Traditional methods for early detection of melanoma rely on the visual analysis skin lesions performed by a dermatologist. The is based so‐called ABCDE (Asymmetry, Border irregularity, Colour variegation, Diameter, Evolution) criteria, although confirmation obtained through biopsy pathologist. proposed method exploits an automatic pipeline morphological and evaluation lesion dermoscopy images. Preliminary segmentation pre‐processing image SC‐cellular neural networks performed, in order to obtain ad‐hoc grey‐level that further exploited extract analytic innovative hand‐crafted features oncological risks assessment. In end, pre‐trained Levenberg–Marquardt network used perform clustering such achieve efficient nevus discrimination (benign against melanoma), as well numerical array be follow‐up rate definition Moreover, authors evaluated combination stacked autoencoders lieu step.