Skin Lesion Segmentation Using Deep Learning

Dermoscopic image analysis has gained significant importance for dermatological applications due to its potential in eliminating observer bias. Segmenting lesion areas automatically in these images is crucial for expedited disease diagnosis. In this study, an innovative deep learning-based approach for precise lesion delineation has been proposed. Leveraging a deep convolutional neural network, where input images are included after appropriate preprocessing. Pedro Hispano Hospital (PH2) 2013 dataset has been employed, comprising dermoscopic images and lesion binary masks as ground truth for evaluation. The approach uses the SegNet architecture, which achieves remarkable accuracy of 99.14%, surpassing benchmark standards (DC 96.94, JI 99.11).This research offers a promising solution for automated lesion segmentation in dermoscopic images, advancing the field of medical image analysis.