Melanoma lesion detection and segmentation using deep region based convolutional neural network and fuzzy C-means clustering.

摘要: Abstract Objective Melanoma is a dangerous form of the skin cancer responsible for thousands deaths every year. Early detection melanoma possible through visual inspection pigmented lesions over skin, treated with simple excision cancerous cells. However, due to limited availability dermatologists, alone has and variable accuracy that leads patient undergo series biopsies complicates treatment. In this work, deep learning method proposed automated region segmentation using dermoscopic images overcome challenges within images. Materials methods A based convolutional neural network (RCNN) precisely detects multiple affected regions in bounding boxes simplify localization Fuzzy C-mean (FCM) clustering. Our constitutes three step process: refinement, region, finally Melanoma. We applied on benchmark dataset ISIC-2016 by International Symposium biomedical (ISBI) having 900 training 376 testing dermatological Main findings The performance evaluated various quantitative measures. achieved average values pixel level specificity (SP) as 0.9417, sensitivity (SE) 0.9781, F1 _ s core 0.9589, (Ac) 0.948. addition, dice score (Di) was recorded 0.94, which represents good performance. Moreover, Jaccard coefficient (Jc) averaged value entire 0.93. Comparative analysis state art results have demonstrated superiority method. Conclusion contrast systems, RCNN capable compute features amen representation Melanoma, hence improves can detect diseases same well different patients efficient mechanism. Series experiments towards validates effectiveness our